diff --git a/.gitattributes b/.gitattributes
index a6344aac8c09253b3b630fb776ae94478aa0275b..e27fe35e7f7a5a6efc6fdb72fa0a8fd398c879e4 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.jsonl filter=lfs diff=lfs merge=lfs -text
+*.tsv filter=lfs diff=lfs merge=lfs -text
diff --git a/README.md b/README.md
index 7b95401dc46245ac339fc25059d4a56d90b4cde5..41653912482ec0b40e1d70a49cadc3e1b03bae7e 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,21 @@
----
-license: apache-2.0
----
+## 代码
+https://github.com/XiaoshanHsj/speechllm/tree/multispk_lingmeng
+
+看代码的README_lingmeng.md
+
+## 数据：
+英文部分: `./data/Whisper-Sidecar-data-metadata/data_for_wavllm`
+
+德文相关: `./data/de-en-mix`
+
+targetASR (target-talker ASR)用到的reference audio: `./data/reference_enroll_audio/all`
+
+英文部分只有metadata，可以从librispeech生成。德文部分还备份了测试集音频。
+
+## 模型：
+tokenizer: `./llama_model/llama/tokenizer.model`
+
+llama-2-chat: `./llama_model/llama-2-7b-chat/consolidated.00.pth`
+
+训练好的MT-LLM模型目录: `./lingmeng_multispk_multitask_retrain_speechllm_v0.1_llama2_chat_wavlm_weighted_update_lora_32_32_prompt_build_multispk_multitask_de.yaml_16gpu_1accum`
+
diff --git a/data/Whisper-Sidecar-data-metadata/convert_to_wavllm_data_format.py b/data/Whisper-Sidecar-data-metadata/convert_to_wavllm_data_format.py
new file mode 100644
index 0000000000000000000000000000000000000000..173d1318dcb07c6abb7c63b88bbce3d38fcc487a
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/convert_to_wavllm_data_format.py
@@ -0,0 +1,42 @@
+import soundfile as sf
+import json
+from tqdm import tqdm
+
+input_jsonl = "/home/v-lingmeng/datasets/Whisper-Sidecar-data-metadata/data/librispeech3mix_test.jsonl"
+output_tsv = "/home/v-lingmeng/datasets/Whisper-Sidecar-data-metadata/data_for_wavllm/" + input_jsonl.split("/")[-1].replace(".jsonl", "1.tsv")
+print(output_tsv)
+head = "\t".join(["id", "audio", "n_frames", "prompt", "tgt_text", "codec", "with_speech", "language", "speakers", "genders"])
+prompts = ['Transcribe the given audio into text. If multiple speakers are speaking, transcribe the utterances of multiple speakers in the order of their start times, separated by "<sc>".']
+
+
+
+with open("/home/v-lingmeng/datasets/LibriSpeech/SPEAKERS.TXT", "r") as f:
+    speaker_info = f.readlines()
+    speaker_gender = {l.split("|")[0].strip():l.split("|")[1].strip() for l in speaker_info if not l.startswith(";")}
+with open(input_jsonl, "r") as f:
+    lines = f.readlines()
+new_lines = []
+for line in tqdm(lines):
+    line = json.loads(line.strip())
+    audio = line['audio']['path'].replace("./dataset", "/valleblob/v-lingmeng/speech/data")
+    wav_id = audio.split("/")[-1]
+    n_frames = str(sf.read(audio)[0].shape[0])
+    prompt = prompts[0]
+    tgt_text = line["sentence"]
+    codec = "None"
+    with_speech = "True"
+    language = "en"
+    if "speakers" in line:
+        speakers = "|".join(line["speakers"])
+    else:
+        speakers = "|".join([_id.split("-")[0] for _id in wav_id.split("_")])
+    genders = "|".join([speaker_gender[spk] for spk in speakers.split("|")])
+    
+    new_line = "\t".join([wav_id, audio, n_frames, prompt, tgt_text, codec, with_speech, language, speakers, genders])
+    new_lines.append(new_line)
+
+with open(output_tsv, "w") as f:
+    new_lines.insert(0, head)
+    f.write("\n".join(new_lines))
+
+# print(speaker_gender)
diff --git a/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_dev.jsonl b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_dev.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..83e81f045cd3345de800d77b0419edec68e127a8
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_dev.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:132bd6f6c6e353bca38831088bbe2eae65ce67635b5cc204249bd070f9c56e2a
+size 2460550
diff --git a/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_test.jsonl b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..16ca7580a2b83fd20ca49bfefd71f71b1c8d7ea9
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53b0d3301937ee8affd9a7da43c8c7940dc6f632b1594efa7de4b7faa4524f5b
+size 1483980
diff --git a/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_test20.jsonl b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_test20.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..16ca7580a2b83fd20ca49bfefd71f71b1c8d7ea9
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_test20.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53b0d3301937ee8affd9a7da43c8c7940dc6f632b1594efa7de4b7faa4524f5b
+size 1483980
diff --git a/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_train.jsonl b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..74f74207936f22f133339863bd01c47d6b41381a
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/aishell1mix2_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a612cd6616cdb756e6c2ad587eebad584e76a1101e754d220bb01b22224c0221
+size 27055103
diff --git a/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_dev.jsonl b/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_dev.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..2f352c8ed15b75a197a85ad6ce05c819b7b0f9a0
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_dev.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bbe7c57288ad1565ffd0d630004508a26b48204b189cf30dd8a0f8f295e01a1
+size 2870838
diff --git a/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_test.jsonl b/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..82b89a70d11a7ff8df5e25bba3572b09635611b2
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2d2fb69a6fbe3cd6c89f89a002e28e01944d96c3d85a6cd240fcb52bfbd8ec2
+size 1442387
diff --git a/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_train.jsonl b/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..b7a8222593fd269b0d63f2cf6371ffad30d7dd42
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/aishell1mix3_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f583f9f8ff84df2ed1eb4da4dfd3a3d615d87c91bd4a3fda34dcf565028e76c
+size 23867370
diff --git a/data/Whisper-Sidecar-data-metadata/data/data_prepare_aishellmix.py b/data/Whisper-Sidecar-data-metadata/data/data_prepare_aishellmix.py
new file mode 100644
index 0000000000000000000000000000000000000000..3a6f0fa5a20a930c64120eb26317d1790df6f633
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/data_prepare_aishellmix.py
@@ -0,0 +1,71 @@
+# prepare from msra-dev-node
+
+import jsonlines
+import soundfile as sf
+import glob
+import numpy as np
+import os
+import pandas as pd
+
+def generate_jsonl_from_fairseq_datafile(root_data_dir, output_dir, with_timestamps=False):
+    splits = ['test', 'dev', 'train']
+    num_spks = ["2", "3"]
+    # convert transcript file to two list, ID and text
+    transcript_path = '/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/aishell1/data_aishell/transcript/aishell_transcript_v0.8.txt'
+    transcripts = open(transcript_path, 'r').readlines()        # ID\ttext\n for each line
+    ID_list = []
+    text_list = []
+    for line in transcripts:
+        line = line.strip()
+        _id, *text = line.split(' ')
+        ID_list.append(_id)
+        text_list.append(''.join(text))
+    
+    id_text_dict = dict(zip(ID_list, text_list))
+
+    for num_spk in num_spks:
+        for split in splits:
+            # data_dir = os.path,join(root_data_dir, 'Aishell1Mix', "data", f'Aishell1Mix{num_spk}', 'wav16k', 'max', split, 'mix_clean')
+            
+            metadata = os.path.join(root_data_dir, 'Aishell1Mix', "data", f'Aishell1Mix{num_spk}', 'wav16k', 'max', 'metadata', f'mixture_{split}_mix_clean.csv')
+            df = pd.read_csv(metadata)
+            mix_id_list = df['mixture_ID'].tolist()
+            mix_path_list = df['mixture_path'].tolist()
+            source_wav_root = os.path.join("/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/aishell1/data_aishell/wav", split)
+            new_jsonl = os.path.join(output_dir, f'aishell1mix{num_spk}_{split}.jsonl')
+            if os.path.exists(new_jsonl):
+                os.remove(new_jsonl)
+            
+            
+            for mix_id, mix_path in zip(mix_id_list, mix_path_list):
+                source_ids = mix_id.split('_')
+                source_texts = [id_text_dict[source_id] for source_id in source_ids]
+                source_text = '</s>'.join(source_texts)
+                
+                speakers = [source_id.split("S")[1].split("W")[0] for source_id in source_ids]
+                
+                source_paths = [os.path.join(source_wav_root, "S"+str(speakers[i]), source_id + '.wav') for i, source_id in enumerate(source_ids)]
+                source_durations = [sf.info(source_path).duration for source_path in source_paths]
+                
+                duration = max(source_durations)
+                dic = {"audio": {"path": mix_path}, 
+                        "language": "zh", 
+                        "duration": duration,
+                        "speakers": speakers,
+                        "sentence": source_text}
+
+                starts = [0] * len(source_durations)
+                ends = source_durations
+                sentences_dict = []
+                for i in range(len(starts)):
+                    sentences_dict.append({"start": starts[i], "end": ends[i], "text": source_texts[i]})
+                dic["sentences"] = sentences_dict
+                
+                with jsonlines.open(new_jsonl, mode='a') as writer:
+                    writer.write(dic)
+
+
+if __name__ == '__main__':    
+    root_data_dir = '/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset'
+    output_dir = '/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/'
+    generate_jsonl_from_fairseq_datafile(root_data_dir, output_dir, with_timestamps=True)
diff --git a/data/Whisper-Sidecar-data-metadata/data/data_prepare_librimix.py b/data/Whisper-Sidecar-data-metadata/data/data_prepare_librimix.py
new file mode 100644
index 0000000000000000000000000000000000000000..085d52d6faf2c4cbbcd3864a075e7e0394a52a7e
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/data_prepare_librimix.py
@@ -0,0 +1,49 @@
+# prepare from SEPC
+
+import jsonlines
+# import soundfile as sf
+import glob
+import numpy as np
+import os
+
+def generate_jsonl_from_fairseq_datafile(root_data_dir, output_dir):
+    splits = ['test', 'dev', 'train']
+    num_spks = ["2", "3"]
+    
+    for num_spk in num_spks:
+        for split in splits:
+            data_dir = root_data_dir + f'Libri{num_spk}Mix_wav16k_max/'
+            wrd =  data_dir + split + '.wrd'
+            fairseq_jsonl = data_dir + split + '_clean.jsonl'
+            new_jsonl = output_dir + f"libri{num_spk}mix_" + split + '.jsonl'
+            if os.path.exists(new_jsonl):
+                os.remove(new_jsonl)
+            
+            with jsonlines.open(fairseq_jsonl) as reader:
+                with open(wrd, 'r') as f:
+                    for meta, text in zip(reader, f.readlines()):
+                        # print(meta, text)
+                        dic = {"audio": {"path": f"./dataset/LibriMix/data/Libri{num_spk}Mix/wav16k/max/" + meta['mixed_wav']}, 
+                            "language": "en", 
+                            "duration": max(meta['durations']),
+                            "speakers": meta['speakers'],
+                            "sentence": text.strip().lower()}
+
+                        starts = meta['delays']
+                        durations = meta['durations']
+                        ends = list(map(lambda x, y: x + y, starts, durations))
+                        texts = text.strip().lower().split(" </s> ")
+                        sentences_dict = []
+                        for i in range(len(starts)):
+                            sentences_dict.append({"start": starts[i], "end": ends[i], "text": texts[i].strip().lower()})
+                        dic["sentences"] = sentences_dict
+
+                        with jsonlines.open(new_jsonl, mode='a') as writer:
+                            writer.write(dic)
+                    
+             
+
+if __name__ == '__main__':    
+    root_data_dir = '/mnt/users/hccl.local/lmeng/workspaces/overlapASR/egs_wav2vec/data/'
+    output_dir = '/mnt/users/hccl.local/lmeng/workspaces/overlapASR/Whisper-Finetune-ovlp/dataset/'
+    generate_jsonl_from_fairseq_datafile(root_data_dir, output_dir)
diff --git a/data/Whisper-Sidecar-data-metadata/data/data_prepare_librispeech.py b/data/Whisper-Sidecar-data-metadata/data/data_prepare_librispeech.py
new file mode 100644
index 0000000000000000000000000000000000000000..b6fdab36f30879d7ee983cec0bf27760a9840246
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/data_prepare_librispeech.py
@@ -0,0 +1,35 @@
+# prepare from SEPC
+
+import jsonlines
+# import soundfile as sf
+import glob
+import numpy as np
+
+def generate_jsonl_from_fairseq_datafile(data_dir, output_dir):
+    splits = ['test']
+    
+    for split in splits:
+        wrd =  data_dir + split + '.wrd'
+        # fairseq_jsonl = data_dir + split + '_clean.jsonl'
+        tsv = data_dir + split + '.tsv'
+        new_jsonl = output_dir + split + '.jsonl'
+        
+        with open(tsv, 'r') as flac_path_f:
+            # remove the first line
+            flac_path_f.readline()
+            with open(wrd, 'r') as trans_f:
+                for flac_path, trans in zip(flac_path_f.readlines(), trans_f.readlines()):
+                    # print(meta, text)
+                    flac_path, duration = flac_path.strip().split('\t')
+                    dic = {"audio": {"path": "./dataset/librispeech/" + flac_path}, 
+                           "language": "en", 
+                           "duration": int(duration)/16000.0,
+                           "sentence": trans.strip().lower()}
+                    with jsonlines.open(new_jsonl, mode='a') as writer:
+                        writer.write(dic)
+
+
+if __name__ == '__main__':    
+    data_dir = '/mnt/users/hccl.local/lmeng/workspaces/overlapASR/egs_wav2vec/data/LibriSpeech/'
+    output_dir = '/mnt/users/hccl.local/lmeng/workspaces/overlapASR/Whisper-Finetune/dataset/'
+    generate_jsonl_from_fairseq_datafile(data_dir, output_dir)
diff --git a/data/Whisper-Sidecar-data-metadata/data/data_prepare_librispeechmix.py b/data/Whisper-Sidecar-data-metadata/data/data_prepare_librispeechmix.py
new file mode 100644
index 0000000000000000000000000000000000000000..738765008dbbbf44b3d5e92130b93544a97df30b
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/data_prepare_librispeechmix.py
@@ -0,0 +1,53 @@
+# prepare from SEPC
+
+import jsonlines
+# import soundfile as sf
+import glob
+import numpy as np
+import os
+
+def generate_jsonl_from_fairseq_datafile(root_data_dir, output_dir, with_timestamps=False):
+    splits = ['test', 'dev', 'train']
+    num_spks = ["2", "3"]
+    for num_spk in num_spks:
+        for split in splits:
+            data_dir = root_data_dir + f'LibriSpeechMix-{num_spk}spkr/'
+            wrd =  data_dir + split + '.wrd'
+            fairseq_jsonl = data_dir + split + '_clean.jsonl'
+            if with_timestamps:
+                new_jsonl = output_dir + f"librispeech{num_spk}mix_timestamps_" + split + '.jsonl'
+            else:
+                new_jsonl = output_dir + f"librispeech{num_spk}mix_" + split + '.jsonl'
+
+            if os.path.exists(new_jsonl):
+                os.remove(new_jsonl)
+            
+            with jsonlines.open(fairseq_jsonl) as reader:
+                with open(wrd, 'r') as f:
+                    for meta, text in zip(reader, f.readlines()):
+                        # print(meta, text)
+                        dic = {"audio": {"path": "./dataset/LibriSpeechMix/" + meta['mixed_wav']}, 
+                            "language": "en", 
+                            "duration": max(list(map(lambda x, y: x + y, meta['durations'], meta['delays']))),
+                            "speakers": meta['speakers'],
+                            "sentence": text.strip().lower(),
+                            }
+                        if with_timestamps:
+                            starts = meta['delays']
+                            durations = meta['durations']
+                            ends = list(map(lambda x, y: x + y, starts, durations))
+                            texts = meta['texts']
+                            sentences_dict = []
+                            for i in range(len(starts)):
+                                sentences_dict.append({"start": starts[i], "end": ends[i], "text": texts[i].strip().lower()})
+                            dic["sentences"] = sentences_dict
+                        
+                        with jsonlines.open(new_jsonl, mode='a') as writer:
+                            writer.write(dic)
+                    
+             
+
+if __name__ == '__main__':    
+    root_data_dir = '/mnt/users/hccl.local/lmeng/workspaces/overlapASR/egs_wav2vec/data/'
+    output_dir = '/mnt/users/hccl.local/lmeng/workspaces/overlapASR/Whisper-Finetune-ovlp/dataset/'
+    generate_jsonl_from_fairseq_datafile(root_data_dir, output_dir, with_timestamps=True)
diff --git a/data/Whisper-Sidecar-data-metadata/data/generate_librimix_wav_from_jsonl.py b/data/Whisper-Sidecar-data-metadata/data/generate_librimix_wav_from_jsonl.py
new file mode 100644
index 0000000000000000000000000000000000000000..7e15ad17148fa84b51342a78150b27c44bafc989
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/generate_librimix_wav_from_jsonl.py
@@ -0,0 +1,60 @@
+import json
+import os
+import glob
+import soundfile
+import librosa
+from tqdm import tqdm
+import numpy as np
+
+def get_delayed_audio(wav_file, delay, sampling_rate=16000):
+    audio, _ = soundfile.read(wav_file)
+    delay_frame = int(delay * sampling_rate)
+    if delay_frame != 0:
+        audio = np.append(np.zeros(delay_frame), audio)
+    return audio
+
+def mix_audio(wav_files, delays):
+    for i, wav_file in enumerate(wav_files):
+        if i == 0:
+            audio = get_delayed_audio(wav_file, delays[i])
+        else:
+            additional_audio = get_delayed_audio(wav_file, delays[i])
+            # tune length & sum up to audio
+            target_length = max(len(audio), len(additional_audio))
+            # print(additional_audio.shape)
+            audio = librosa.util.fix_length(audio, size=target_length)
+            additional_audio = librosa.util.fix_length(additional_audio, size=target_length)
+            audio = audio + additional_audio
+    return audio
+
+
+jsonl_path = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/libri3mix_train.jsonl"
+if "test" in jsonl_path:
+    subset = "test"
+elif "train" in jsonl_path:
+    subset = "train"
+else:
+    subset = "dev"
+librispeech_dir = f"/home/v-lingmeng/datasets/LibriSpeech/{subset}*/"
+output_dir = "/home/v-lingmeng/datasets"
+
+
+with open(jsonl_path, 'r', encoding='utf-8') as file:
+    json_list = [json.loads(line.strip()) for line in file]
+
+for line in tqdm(json_list):
+    audio_name = line['audio']['path'].replace("./dataset", output_dir)
+    if not os.path.exists(os.path.split(audio_name)[0]):
+        os.makedirs(os.path.split(audio_name)[0])
+    
+    source_ids = os.path.split(audio_name)[1].split(".")[0].split("_")
+    source_files = [glob.glob(librispeech_dir + "/".join(i.split("-")[:-1]) + f"/{i}*")[0] for i in source_ids]
+    delays = [l["start"] for l in line['sentences']]
+    # ends = [l["end"] for l in line['sentences']]
+    
+    # for source_audio, start, end in zip(source_auidos, starts, ends):
+    mixed_audio = mix_audio(source_files, delays)
+    soundfile.write(audio_name, mixed_audio, samplerate=16000)
+    print(audio_name)
+    
+# print(delays)
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_dev-both.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_dev-both.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..83b65d65629a73073f8fb593751ba4405c25194d
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_dev-both.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a00333b0e46f84e4389e63df5045f8a762487eb29bb89eb7de9f70941e1434c
+size 2224222
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_dev.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_dev.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c1eee09ef5626ef4b7c435d05977f6dfdf151fbe
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_dev.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c08740dd203b33312ea9793a8859302c2ff41f44de77ffb404c6d5859df2287
+size 2227222
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_test-both.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_test-both.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..6baa9637f0698ae31ed313a5a0523ce84bd61fce
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_test-both.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88e089c638beff65237045f526348e50341026b0324e0230e1ee17568b3173c1
+size 2094572
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_test.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..a47d98dafb58428c7795ec3a0d811aa69809c19c
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbf70cb2bca21af6465f534a1910756cce13eb410815eccc90fc74b24567e4b6
+size 2097572
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_test20.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_test20.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c4d135d44aff07cf1778f12556f12c568c1eb187
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_test20.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9831f61da77ba9a17367845c6288c9a7b24858cf69dcbb53e4377cb23640820b
+size 14417
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-100-both.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-100-both.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..f90410f9a56f116be4dcc06fc122f73bff438787
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-100-both.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f0eb2462fa951da717d18e172a5b781b6149d3483e56022bcea7459d163b76e
+size 14778639
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-100.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-100.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..441f03f6b1c925766eb2c33d33474aebc7b46abf
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-100.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10ac8db4f8c3851c7e34f4643268355c2d34a7f8952aa316b0b70d2019733f70
+size 14792539
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-200.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-200.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..d19d9a5d339f7e1d6243bea87e2ebd0c7ec8be89
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-200.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94e70011854f573bf05586a9793f19510d6be7188be89c032459910ad1d3e11d
+size 29571178
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-both.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-both.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e7968254be4f1bbccdef8adfbc65e2ad78859bbc
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train-both.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f72957a086f85d8fee3aea616c471dd7c7baf465d355db361f796f13c7e6478e
+size 68694922
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_train.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8e23d8c7e8a0da190244b386c1e5a6222f04b1dc
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9bd7eceb100a8f842e1b708066e5da7ac77e5008ef013266c567528ab1f8c4f
+size 68759621
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_train20.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train20.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0d741f2a61c4fcbb37a5d19b20973cd701343acf
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train20.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41d2502e124c4cdf60adb5a83f8942f2fb5d710634485b5a0382e45873570048
+size 22316
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri2mix_train_remove_enroll.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train_remove_enroll.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..826701e00a17189a20e0867eb83a85cc21c19af3
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri2mix_train_remove_enroll.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3d3c34060e44c767cd70ebb53aa18d53db4a3828fa92037addeed9c04e9f8a5
+size 68268344
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri3mix_test.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri3mix_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..23a0a11ce26a1c669db47883a503b35b785bec87
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri3mix_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daf3f6411bb717e86d77b4f17a455323c5d9df5fdb37fe6513a11859d50f41e9
+size 2855611
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri3mix_test20.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri3mix_test20.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..07e5e0ea13d078ac93f23952be8a308cf6d668c2
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri3mix_test20.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f66ad5735f3e3d1bd24bf960439142e181dbe4ab50a951a3807b4050040568c
+size 19612
diff --git a/data/Whisper-Sidecar-data-metadata/data/libri3mix_train.jsonl b/data/Whisper-Sidecar-data-metadata/data/libri3mix_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c3afe0e0834935b8c65279c6474eb36636fee9ba
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/libri3mix_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ba19b8a28a4430dbab099d5450ae075f6096c869ca318846f1ed219b471c997
+size 65232138
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_test.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..475876d0b322c33cdc4731b37ed3ed8aea32f418
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6dfd13ad50aeb92257fc7c8c4d721fad1d2e3901be95eed0aa4d58211027d471
+size 1945661
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_test_30s.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_test_30s.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..9ab4cb9957c6628322651c187063360b76d63b98
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_test_30s.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1d07d099fe7d64cfec16efcc06b5dcac825984ca8668909152882adef2edcb9
+size 1945138
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_train.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..bdf4685fa7abc23e3d97a9db791fb0995d608dd6
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech2mix_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f629fb7f89d02d937e5f4d1ced117c70904fc5d97d520706758a3e41010c1c2
+size 287742080
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_test.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..5940d92a012f113d43c1cbe2c775fff75a2c1ada
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4019227f5a2f827f2517e942c8c6f72b7a6d80fe8fe32db2909393254d7af771
+size 2730857
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_test_temp.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_test_temp.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..5399b2d7979284e66140c07c0d6c3bc61cb2cd9b
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_test_temp.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be1fafb707b168a23d3c8225e3f04407da63360b27b99d435d11c8a7560dee9c
+size 2729249
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_train.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0c145ab4c837c491785c00d860a2c3738c1f40b7
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech3mix_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47917569ed5a933e6ad66ccef5ccefded8f4c117a0296dafcbde9fc564e835e6
+size 410988677
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech_dev.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech_dev.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8e1bb7c08abc3a7a164d0ca915366f11fe5aad52
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech_dev.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca54eb22dd7f421bd9c8b8e60a5df32946a7823d5547e49eaa030769c48de0da
+size 2194617
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech_test.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e22bcd6c7a34d74b15ff8eab829cc5b691e809ad
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8421fbe45b046b2a5644782b3e982538bd8483d72964b6f08fc208e5e7059648
+size 2197082
diff --git a/data/Whisper-Sidecar-data-metadata/data/librispeech_train.jsonl b/data/Whisper-Sidecar-data-metadata/data/librispeech_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..3698674e4af767643af27d1f1cd77099cd008a42
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/librispeech_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc923ead53a1248037e622bbeb75899a0c2637720600cc8c386530d5cdfa95b8
+size 91088243
diff --git a/data/Whisper-Sidecar-data-metadata/data/long_wav_resample.py b/data/Whisper-Sidecar-data-metadata/data/long_wav_resample.py
new file mode 100644
index 0000000000000000000000000000000000000000..d0dd08ca20c1e87dd4d66a6b26e4dcd729c4e994
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/long_wav_resample.py
@@ -0,0 +1,52 @@
+# resamle long wav (>30s) to 16k 30s, and update the jsonl file
+
+
+import jsonlines
+import os
+
+from pydub import AudioSegment
+from pydub.playback import play
+import soundfile as sf
+
+
+
+jsonl_file = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/librispeech3mix_test.jsonl"
+temp_file = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/librispeech3mix_test_temp.jsonl"
+# time strench long wav (>30s) to 16k 30s, and update the jsonl line in-place
+
+with jsonlines.open(jsonl_file, 'r') as reader, jsonlines.open(temp_file, 'w') as writer:
+    for obj in reader:
+        wav_path = obj['audio']['path']
+        duration = obj['duration']
+        sentences = obj['sentences']
+        # resample wav
+        if duration > 30.1:
+            print(wav_path, duration)
+            wav = AudioSegment.from_file(wav_path)
+            target_len = 30.0 * 1000
+            speed_up_rate = len(wav) / target_len
+            wav = wav.speedup(playback_speed=speed_up_rate)
+            wav = wav[:target_len]
+            wav.export(wav_path, format="wav")
+            print(speed_up_rate)
+            obj['duration'] = 30
+            for sentence in sentences:
+                sentence['start'] = sentence['start'] / speed_up_rate
+                sentence['end'] = sentence['end'] / speed_up_rate
+                if sentence['start'] > 30:
+                    sentence['start'] = 30
+                if sentence['end'] > 30:
+                    sentence['end'] = 30
+            obj['sentences'] = sentences
+        elif duration > 30.0:
+            wav, sr = sf.read(wav_path)
+            wav = wav[:int(16000 * 30)]
+            obj['duration'] = 30 if duration > 30 else duration
+            for sentence in sentences:
+                sentence['start'] = sentence['start'] if sentence['start'] < 30 else 30
+                sentence['end'] = sentence['end'] if sentence['end'] < 30 else 30
+            obj['sentences'] = sentences
+            sf.write(wav_path, wav, sr)
+
+        
+        writer.write(obj)
\ No newline at end of file
diff --git a/data/Whisper-Sidecar-data-metadata/data/select_prompt_wav.py b/data/Whisper-Sidecar-data-metadata/data/select_prompt_wav.py
new file mode 100644
index 0000000000000000000000000000000000000000..3eacbb47281212d29581296a59dfe9324214c647
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/select_prompt_wav.py
@@ -0,0 +1,132 @@
+import os
+import glob 
+import random 
+
+import jsonlines
+
+# backup_dir = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/temp/removed_mix_wav"
+# librispeech_dir = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/LibriSpeech/train-clean-360"
+# enroll_dir = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/temp/new_enroll"
+
+# # 1. 获取所有的wav文件
+# wav_files = glob.glob("/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/LibriMix/data/Libri2Mix/wav16k/max/train-360/mix_clean/*.wav")
+
+# # 2. 每个wav文件有两个说话人，记录所有wav文件出现的说话人
+# all_speakers = set()
+# for wav_file in wav_files:
+#     speakers = [f.split('-')[0] for f in os.path.basename(wav_file).split("_")]
+#     all_speakers.update(speakers)
+# # random.shuffle(wav_files)
+
+# len_all_speakers = len(all_speakers)
+# # 3. 对每个说话人，复制且只复制一个具有它的语音文件
+# count = 0
+# for wav_file in wav_files:
+#     source_wavs = os.path.basename(wav_file).split("_")
+#     speakers = [f.split('-')[0] for f in source_wavs]
+#     # 如果有任意一个说话人不在all_speakers中，跳过这个文件
+#     if not all(s in all_speakers for s in speakers):
+#         continue
+#     else:
+#         # 从all_speakers中删除这两个说话人
+#         all_speakers.difference_update(speakers)
+#         # 复制这个文件
+#         os.system(f"cp {wav_file} {backup_dir}")
+#         # print(f"cp {wav_file} {backup_dir}")
+#         # 复制source_wavs中的每个说话人的语音文件
+#         for source_wav in source_wavs:
+#             count+=1
+#             source_wav_path = os.path.join(librispeech_dir, source_wav.split("-")[0], source_wav.split("-")[1], source_wav.split('.')[0] + ".flac")
+#             # 判断是否存在
+#             if not os.path.exists(source_wav_path):
+#                 print(f"source_wav_path: {source_wav_path} not exists")
+#                 continue
+#             os.system(f"cp {source_wav_path} {enroll_dir}")
+            
+# print(all_speakers)
+# print(count, len_all_speakers)
+
+
+
+# source = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/temp/new_enroll"
+# enroll_dir = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/LibriMix_enroll_audio/train-360"
+
+# flac_files = glob.glob(f"{source}/*.flac")
+
+# for flac_files in flac_files:
+#     # mkdir enroll_dir/spk_id
+#     spk_id = os.path.basename(flac_files).split("-")[0]
+#     spk_dir = os.path.join(enroll_dir, spk_id)
+#     if os.path.exists(spk_dir):
+#         # 删除
+#         os.system(f"rm -rf {spk_dir}")
+#     os.makedirs(spk_dir, exist_ok=True)
+#     # convert flac to wav, move to spk_dir
+#     wav_file = os.path.join(spk_dir, spk_id+ ".wav")
+#     os.system(f"ffmpeg -i {flac_files} {wav_file}")
+    
+
+# jsonl_file = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/libri2mix_train_remove_enroll.jsonl"
+# enrolled_dir = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/temp/removed_mix_wav"
+
+# # remove wav file in enrolled_dir from jsonl_file
+# with jsonlines.open(jsonl_file) as reader:
+#     lines = list(reader)
+#     print(len(lines))
+#     for line in lines:
+#         mix_wav = line['audio']['path']
+#         if os.path.exists(os.path.join(enrolled_dir, os.path.basename(mix_wav))):
+#             lines.remove(line)
+#     print(len(lines))
+# # write to new jsonl file
+# new_jsonl_file = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/libri2mix_train_remove_enroll.jsonl"
+# with jsonlines.open(new_jsonl_file, "w") as writer:
+#     for line in lines:
+#         writer.write(line)
+# print("done")
+
+
+librispeech_dir = "/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/LibriSpeech/train-other-500"
+enroll_path ="/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/LibriMix_enroll_audio/train-500"
+
+# 检查librispeech_dir中每个speaker是否存在在enroll_path中
+speaker_dirs = glob.glob(f"{librispeech_dir}/*")
+new_file = []
+for speaker_dir in speaker_dirs:
+    if ".TXT" in speaker_dir:
+        continue
+    speaker_id = os.path.basename(speaker_dir)
+    enroll_speaker_dir = os.path.join(enroll_path, speaker_id)
+    if not os.path.exists(enroll_speaker_dir):
+        print(f"{enroll_speaker_dir} not exists")
+        os.makedirs(enroll_speaker_dir, exist_ok=True)
+
+    if len(glob.glob(f"{enroll_speaker_dir}/*.wav")) == 0:
+        # 从librispeech_dir中复制一个语音文件到enroll_speaker_dir
+        flac_files = glob.glob(f"{speaker_dir}/*/*.flac")
+        # 从flac_files中随机选择一个, 并复制到enroll_speaker_dir
+        flac_file = random.choice(flac_files)
+        # 复制到enroll_speaker_dir
+        new_flac_file = os.path.join(enroll_speaker_dir, os.path.basename(flac_file))
+        print(new_flac_file)
+        os.system(f"cp {flac_file} {new_flac_file}")
+        # 记录flac_file的名字
+        new_file.append(flac_file)
+    else:
+        print(glob.glob(f"{enroll_speaker_dir}/*.wav"))
+    
+    # 检查enroll_path中speaker的语音命名是否为{speaker_id}.wav
+    enroll_wav_files = glob.glob(f"{enroll_speaker_dir}/*")
+    for enroll_wav_file in enroll_wav_files:
+        if os.path.basename(enroll_wav_file).split(".")[0] != speaker_id:
+            print(f"{enroll_wav_file} not match")
+            # 转为wav格式, 名字只保留speaker_id
+            wav_file = os.path.join(enroll_speaker_dir, speaker_id + ".wav")
+            os.system(f"ffmpeg -i {enroll_wav_file} {wav_file}")
+            # 删除原来的文件
+            os.system(f"rm -rf {enroll_wav_file}")
+
+# 记录新的flac文件
+with open("/home/v-lingmeng/codebase/Whisper-Finetune-ovlp/dataset/temp/new_flac_files.txt", "w") as f:
+    for flac_file in new_file:
+        f.write(flac_file + "\n")
\ No newline at end of file
diff --git a/data/Whisper-Sidecar-data-metadata/data/test_examples.jsonl b/data/Whisper-Sidecar-data-metadata/data/test_examples.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..2849ffb4362d2ce19b8d8d38b84fe0805c1c0fc2
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data/test_examples.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcf048b062b475972c1e6dad258244680a88a9acfc8841dd8ea284ad56efc3dc
+size 794
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..4e331b020395b5af712ff4de4b05c6bc6af5d945
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53fb93ebd24db3e32a7dd7fd5908222f2f0a498a7d0ede6835140c2d4b1f3552
+size 8892480
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_1350.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_1350.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..b85f5e8558a95858275292f9cb6f65146fc93ea5
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_1350.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec64b8b840655de399e374931c056dd4b22d4070c8b97cb197863182599af765
+size 889784
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_1350_targetLingual.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_1350_targetLingual.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..312cc13add57e5b63a0240911efd2f1140d9b559
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_1350_targetLingual.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:971baf65ee1603b22483ef12a1040d884419077069b68dad7931a305ffe2f7aa
+size 632636
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_targetLingual.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_targetLingual.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..c8e6e1a99cd79d9679cc0a2fe11a04520506aef6
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_targetLingual.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12249076735100be44176352c65ab89a30751ae7ac5621583130674529774d57
+size 6361360
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_targetLingual_1350.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_targetLingual_1350.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..b4a4679c8eea26a376f6a8c6dd2a71f46f52036e
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_targetLingual_1350.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:968fd7a78bdca0783e5a2baf7db53f621ed274ba1295f5bd4d2553794a4718da
+size 636201
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..620b99f0cb43303f40ba78427fe9739463e79885
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21cb5e6b2243c0ce5a2cf6e431dd1c9c5dc5efb79fb3c2a9d548dbbc335c212f
+size 93936028
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train_targetLingual.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train_targetLingual.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..a400badf704aa3ec62c2c14c994046325c8c108c
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train_targetLingual.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8aad7b8718ed09f7e98742caa6c85e496f68e332f73e65a557ecdac06cec8f2a
+size 65029852
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..1e95569820361328578d040ee3a172938f3f6fba
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2e84678170247fd7ef2a06622a32d9d42abf3e1b83e8c4426e4a9d03e44ce72
+size 10851471
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_1350.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_1350.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..67193a21e8df640db4654367ca52a17e12248da5
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_1350.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aaa506c275e319dccec8357afa545cb04a77fbe8db48579248546658351c2590
+size 1085187
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_1350_targetLingual.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_1350_targetLingual.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..7cf60b20efad4d614a12f9d4666375964d80ec01
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_1350_targetLingual.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d01a55200a1413693e8c8bb5d24ed36e11e56a8a49f1607613fc1f6b9bc3120
+size 759724
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_targetLingual.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_targetLingual.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..1910051665cf56ff82fe3c94f9dfbbf66bcd0946
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_targetLingual.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c109b3299a7540bb54a336176f0b5bf14ae719a79270b674561786dcd3d2fd2c
+size 7570211
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_targetLingual_1350.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_targetLingual_1350.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..bcd04743241d84cc93fbe85f79f788092a95cc92
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_targetLingual_1350.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc4e78e235c28b2942eb1a198e97f19aa4ac2c8bbf134fee2c72cc4740bfe44e
+size 757435
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..cb6bed6f2e05799e8d95b5c165958f4d006e87f7
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9c5519a7a236abf99102ee6b29291494e4dcb2af55d343e9db1db106ed19455
+size 122416553
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train_targetLingual.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train_targetLingual.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..c71d25f0ec4677f3b3217bdede392f15d931bd37
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train_targetLingual.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4eae93d3edada6ce3273b5b37561a76575f157b16070c39ad071859e206d3bc
+size 81906653
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..341bbab03dd9583f4973fc8d9f4fcfbaa53cef9a
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f7251d28b23392256aaa607566237762247ff81fa40dd5a90ae6d780a47b41c
+size 5227950
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_test_1350.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_test_1350.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..c646302b7da3836ffbf0cf422b3b51e5772c7f88
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_test_1350.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9b0220a62885edee96adcd238931abd7ae2a4af288ee1e0d15bc67b0685ddbf
+size 522097
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..36e2d35545edf13c0cb8654a75c9f386240d7cf2
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3911cbb91e28758c7380c6987c5896f00f0e46308b6fc5a7207d521d65265692
+size 48866960
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/dict.txt b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/dict.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d328f2152d8b37a96b3e5470c56fcd6755a7a7ad
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/dict.txt
@@ -0,0 +1,5 @@
+1 1
+2 2
+3 3
+4 4
+5 5
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/fix_data.py b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/fix_data.py
new file mode 100644
index 0000000000000000000000000000000000000000..e99bbb922d3d39c1fdfa80a4a1bdfced4e97275b
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/fix_data.py
@@ -0,0 +1,12 @@
+# change the separation symbol from <\s> or </s> to <sc>
+
+import glob
+
+for filename in glob.glob('*.tsv'):
+    with open(filename, 'r') as f:
+        lines = f.readlines()
+    with open(filename, 'w') as f:
+        for line in lines:
+            line = line.replace('</s>', '<sc>')
+            line = line.replace('<\s>', '<sc>')
+            f.write(line)
\ No newline at end of file
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..dfd915af128477b0e041338beffb4444e5288bc0
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abd782476290544d5874dadeafa39b8aa215f2c415a2fe1f77bd6cfdb9a4d7d4
+size 1707193
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..a61cb05f48e8d8214376c8983147335d99d17469
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cd2227e6677c0d60d1d3c98e179dc800ee6688785ac04cb9c4f92ef137f3a52
+size 554102
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..2a1c019db2337a1a038e2e89c5534895c00287ca
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac66eef4cf3828a4d6947dd98bbe838a3b440b73255117385d001017861197f4
+size 1215967
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..9377fd4f83000e061fad98f1c941df1ea1ed8a37
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c3edb59b3c35638832449c9c359cdb35d758fef61d9e93413f7fe04593fce47
+size 1731193
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..c0748346b5c4bbae88d4e5396bf37955876831ed
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5d20598993fee48384704a561940f32213e50136d7e67c13446255c4d456a27
+size 48753757
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..e68d701ee681eb66e7fd729ddbad92e44ee37330
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a07645227eb5c5b1c2079d9383fc972339fe41faedf5ec8985227c85fb8f88ef
+size 14997756
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..80ddbb7f7e712ace45c35874d928794cc747061d
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ae5a2c6bca8656559e2a1df923c5c675e7e880fdae1252e83ecfe1730dd7307
+size 32300627
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..fa2d8d5b23ca5ee52a4e6aaf4d4dac09640da3d5
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:406b6cf55a146f26de8fcd39f997c059b80c24a87190cdb8a46182383b6b7fad
+size 49271349
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..43471d732706273d89753b1178d54b16df7f7cb4
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58a4e9e2713d4edfa639f7c948c44dd7473246538044487a21f27bb37e9c5057
+size 2122593
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..1c14598794235e5fe8505c9c74f908cfeb305b47
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:580a5ec72ecea86a4cdbf4289552033f699d351ae9b4b28c9871df2b8d15bf73
+size 1043335
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..5a46772be39c9062e3610b3a7ba46bc14de8fa78
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b069534584568efe90d3c90f522a5423cb1b0196ef43774cdfecdf8e539818f
+size 1335726
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..7148a2add22aa6777797235c3e8c43aad0c8e134
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:866e8cd48cbc4a1e4f4bdbbe47677ef26388887b76cee1b98e9ab4d40c37cc7d
+size 2137593
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..86f7be4c7213a83220956c8274ccaefdf2e6101a
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ad0fbaba6567615ed2e71a4fd123800d0c1687103f7cdcea54d5e3147de1581
+size 42611054
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..f9d33a32bc1a931de06c3ec60bb97fdb0e11114d
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84151ac531719332cc40c810748f5b1f5618d5d5a5ebdb3df869f0057ebcf85a
+size 19381505
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..2a0ae6d92b15628594f8633d097c5bb8c257fd2f
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0701d55a886c38a652798761074c945cc0b7c49266e514324057436f30617f17
+size 23397918
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..886c976bcb9033414b84f17902c7ba4bbd447ce3
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45b552729e38a4f68730dcf254e892228bd5a28d9482269769319592dbc42c3c
+size 42956654
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..19d84c7680fc927d96efdba74b2afa6597ed4934
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a7da83295679a995a07db1775afe73e820db8d0e4a7b2a8fa4cfb7b44694d54
+size 1429201
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..4df887bf6aea65b4e9ac9229b25b01389889f7fb
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa83c5427d1658509f7c88476a93a8e71c6b28e06f91c2364c2ce0999b7c30ba
+size 443151
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..240f3ea716fbb81154dd96f8ce8732b1df395619
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ecdb004f7f813b33febd551fa047f14c51ea79db38b9e983dbc4538bcc5575c
+size 1002591
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_order.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_order.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..38023789179139cb20ce8524f4239dbb4ff8563a
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_order.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efa60e033db5970c6d7d7dd8213322a685b528430c6f622ffb29e1e148f0c05d
+size 939769
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..c24ab598b0f838f79ec40969b26d18de4ebb7ed3
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90cb5be6066df8f49d1a60d52b386887e4738bfcc7f297c28355ffc299c68e3a
+size 1450145
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..cd3eeeb428e2a29e65b7dbba2c2edf969d25641b
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bc631089c416cb1f8f85cb7fac368fc2154ad881821103ab6168c7200e40914
+size 193134514
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..1cc74f859fcb822624d7c89bc7fbc646722b4d92
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a39a5e94f1dc225559717e5ea18630824f9fb55a4ed893b45a0d619fefd790f
+size 57154620
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..fcda79e59427e3260ba3e3db3bdd98ddfce90f21
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8e3a4235b3d8911f38e6dda8d600fa5731fa94156fe3eea12e91d1982cab43d
+size 125481936
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_order.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_order.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..5b58d45523414cc2fc7385fbb6224c105e38565d
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_order.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d3e376e9f8861738f6658f1e3a46f3143a3be0693369d9977d4ae37d43563c7
+size 121260123
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..5d0c54402fe802c33b24f0b7e9c31332b76c2204
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4020aa3de361138ac6aa9e6b2bda5d5b0a9a5435d67ee85996afc003eedba08
+size 195384426
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..40b3851bd751fbfa40066761c8372ef97de05269
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:315e1c33aeeaaf8e15e6880c23b6fd0c7d3b751b9edf512aa7f83dd9fd8defec
+size 1726904
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..6c309aeee035a28a47228a0c4609d525229c0f3c
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f736fe7baf5c4d74dba9814c54c3a59781f0d20a0e4f97095908080ec55d1a84
+size 805156
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..22cdd96afcafbe1a7c571da4320643f49098df3a
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54bd8b064f2a4367f8703ae2b48bbb7d67294a9fb9e7b48fa28dce14d71be84b
+size 1037649
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_order.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_order.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..5957d33591d937d7cdac1da3085b608d25f76c78
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_order.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e476b6d9164638f8f4fb985663518963131a80da9e5798e1ca74c410d007d047
+size 955658
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..f74b36fe825b2a9d69c4f762af6d7458e1217df7
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1408cd311290ea46bb706ce83b277b74371296cdc1fbbddc5529f0d49955eef3
+size 1747728
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..038f7c89825d5ff80569b5396d228a28a3eb5d28
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5beee364fb409e50032387cd6695504c342c2b7005fb0be7685281f70a7e41e8
+size 242713442
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_gender.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_gender.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..bf51fe311ad804fb830b44168062258b9350d1db
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_gender.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d50912112bea06dffcb59c00ea8d429b3ccce7f121fdffeba94cc51c0c7a944b
+size 104772496
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_keyword.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_keyword.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..549a6da65ef2dfe0d114dbf3eb9ad7e4b2aa2660
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_keyword.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffeb0f22aa8cdf06d49bf3b8758574802fbab3783e9aa0afd96421091b57cca2
+size 126509909
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_order.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_order.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..f4cc662479bc38603cf93e6fc24d446c6ef24eb0
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_order.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5abe1fe2b83fec9db0494715ffa3ca573641103ee630e577384ad51c943d1f1c
+size 121342102
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_targetASR.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_targetASR.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..654258796fdfe40fffe3236317df9d4ea86e9db9
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_targetASR.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94576fbc9162c6e664c1fd3a01a9a9d8327e08b59dae32b61594ad7132bf5f04
+size 244932674
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_test.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_test.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..5492d510cc55e72cf7d8813d13bcd3824819cb3a
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_test.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0753839e606effdc7b480b00adbb5622dbf067d62c9e70f7664215f0a44afbc
+size 2366608
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_test_clean.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_test_clean.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..1e8e1fb140f763d5666f4958cf7cba439a1203b0
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_test_clean.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d6608467e6bbe61d3459e60233f75c26be62fc29eb2e20d43d4f9ebbc192855
+size 1133930
diff --git a/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_train.tsv b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_train.tsv
new file mode 100644
index 0000000000000000000000000000000000000000..857d1e829e9d565d88561277eb0355bd5201a0e9
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_train.tsv
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55fac030be50f6c70fe72089cd75cd5b6916395a5d626ff109e075b57cd3d656
+size 143035279
diff --git a/data/Whisper-Sidecar-data-metadata/generate_multitask_data.py b/data/Whisper-Sidecar-data-metadata/generate_multitask_data.py
new file mode 100644
index 0000000000000000000000000000000000000000..3cef92ff7923c07246940afcc72f703e5d75b829
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/generate_multitask_data.py
@@ -0,0 +1,141 @@
+import glob
+import random
+
+tsv_files = glob.glob("/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/*.tsv")
+
+
+
+head = "\t".join(["id", "audio", "n_frames", "prompt", "tgt_text", "codec", "with_speech", "language", "speakers", "genders"])
+for tsv in tsv_files:
+    gender_lines = []
+    order_lines = []
+    keyword_lines = [] 
+    target_lines = [] 
+    targetLingual_lines = []
+    if "gender" in tsv or "order" in tsv or "keyword" in tsv or "librispeech_" in tsv or "target" in tsv:
+        continue
+    with open(tsv, "r") as f:
+
+        _ = f.readline()
+        lines = f.readlines()
+        # # gender task
+        # for line in lines:
+        #     genders = line.split("\t")[-1].strip().split("|")
+        #     targets = line.split("\t")[4].split(" <sc> ")
+        #     if "M" not in genders or "F" not in genders:
+        #         continue
+        #     else:
+        #         target_gen = "male" if random.random() > 0.5 else "female"
+        #     prompt = f"Please transcribe the contents spoken by {target_gen} speakers in overlapping speech."
+
+        #     new_targets = [target for gender, target in zip(genders, targets) if (gender == "F" and target_gen == "female") or (gender == "M" and target_gen == "male")]
+        #     new_targets = " <sc> ".join(new_targets)
+
+        #     new_line = line.strip().split("\t")
+        #     new_line[3] = prompt
+        #     new_line[4] = new_targets
+        #     new_line = "\t".join(new_line)
+        #     gender_lines.append(new_line)
+        # output_file_gender = tsv.replace(".tsv", "_gender.tsv")
+        # with open(output_file_gender, "w") as f:
+        #     f.write(head + "\n" + "\n".join(gender_lines))
+            
+        # # order task
+        # for line in lines:
+        #     if "2mix" in tsv:
+        #         num_spk = 2
+        #     elif "3mix" in tsv:
+        #         num_spk = 3
+        #     else:
+        #         continue
+        #     target_idx = random.randint(0, num_spk-1)
+        #     targets = line.split("\t")[4].split(" <sc> ")
+        #     new_targets = targets[target_idx]
+
+        #     _ = ['first', "second", "third"][target_idx]
+        #     prompt = f"There are multiple speakers in the audio. Please transcribe the speech of the {_} speaker into text."
+
+        #     new_line = line.strip().split("\t")
+        #     new_line[3] = prompt
+        #     new_line[4] = new_targets
+        #     new_line = "\t".join(new_line)
+        #     order_lines.append(new_line)
+        # output_file_order = tsv.replace(".tsv", "_order.tsv")
+        # with open(output_file_order, "w") as f:
+        #     f.write(head + "\n" + "\n".join(order_lines))        
+        
+
+        # # keyword task 
+        # for line in lines:
+        #     # 获取 targets 并将它们拆成单词集合
+        #     targets = line.split("\t")[4].split(" <sc> ")
+        #     # 筛掉长度小于等
+        #     targets_list = [set(word for word in t.split(" ") if len(word) >= 6) for t in targets]
+        #     # 求所有 targets 中的单词集合并
+        #     all_word_set = set.union(*targets_list)
+        #     if len(all_word_set) == 0:
+        #         continue
+        #     # 找出每个 target 中独有的单词
+        #     unique_targets_list = []
+        #     for target_set in targets_list:
+        #         other_targets_union = set.union(*[s for s in targets_list if s != target_set])
+        #         unique_words = target_set - other_targets_union
+        #         unique_targets_list.append(unique_words)
+
+        #     all_unique_word_set = set.union(*unique_targets_list)
+        #     if len(all_unique_word_set) == 0:
+        #         continue
+
+        #     sampled_word = random.choice(list(all_unique_word_set))
+        #     # 找到这个单词属于的集合索引
+        #     for i, unique_words in enumerate(unique_targets_list):
+        #         if sampled_word in unique_words:
+        #             set_index = i
+        #             break
+            
+            
+        #     new_targets = targets[set_index]
+        #     prompt = f'Please transcribe the speech of the speaker who said the word "{sampled_word}" in the overlapping speech audio.'
+
+        #     new_line = line.strip().split("\t")
+        #     new_line[3] = prompt
+        #     new_line[4] = new_targets
+        #     new_line = "\t".join(new_line)
+        #     keyword_lines.append(new_line)
+        # output_file_keyword = tsv.replace(".tsv", "_keyword.tsv")
+        # with open(output_file_keyword, "w") as f:
+        #     f.write(head + "\n" + "\n".join(keyword_lines)) 
+
+        # # target talker ASR
+        # this part is processed by the model dataloader
+        # for line in lines:
+        #     prompt = "The audio file starts with a 3-second reference speech by the target speaker, followed by overlapping speech. Please transcribe the target speaker's part from the overlapping section."
+        #     new_line = line.strip().split("\t")
+        #     new_line[3] = prompt
+        #     new_line = "\t".join(new_line)
+        #     target_lines.append(new_line)
+        # output_file_targetASR = tsv.replace(".tsv", "_targetASR.tsv")
+        # with open(output_file_targetASR, "w") as f:
+        #     f.write(head + "\n" + "\n".join(target_lines)) 
+            
+
+        # target talker lingual
+        if "de-en-" not in tsv:
+            continue
+        for line in lines:
+            target_lingual = random.choice(["English", "German"])
+            prompt = f"Please transcribe the person speaking {target_lingual} from the overlapping speech audio."
+            new_line = line.strip().split("\t")
+            new_line[3] = prompt
+            
+            target_lingual = "en" if target_lingual=="English" else "de"
+            tgt_texts = new_line[4].split(" <sc> ")
+            langs = new_line[7].split("|")
+            tgt_text =  [text for text, lang in zip(tgt_texts, langs) if lang == target_lingual]
+            new_line[4] = " <sc> ".join(tgt_text)
+            
+            new_line = "\t".join(new_line)
+            targetLingual_lines.append(new_line)
+        output_file_targetLingual = tsv.replace(".tsv", "_targetLingual.tsv")
+        with open(output_file_targetLingual, "w") as f:
+            f.write(head + "\n" + "\n".join(targetLingual_lines)) 
\ No newline at end of file
diff --git a/data/Whisper-Sidecar-data-metadata/select_prompt_wav.py b/data/Whisper-Sidecar-data-metadata/select_prompt_wav.py
new file mode 100644
index 0000000000000000000000000000000000000000..fcc666b085a6f23197efc0ae7940d6082ff91a7e
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/select_prompt_wav.py
@@ -0,0 +1,50 @@
+import os
+import glob 
+import random 
+
+import jsonlines
+
+librispeech_dir = "./dataset/LibriSpeech/*"
+enroll_path = "./prompt_enroll_audio/all"
+
+if not os.path.exists(enroll_path):
+    os.makedirs(enroll_path)
+
+speaker_dirs = glob.glob(f"{librispeech_dir}/*")
+new_files = []
+for speaker_dir in speaker_dirs:
+    if not "train-" in speaker_dir and not "test-" in speaker_dir and not "dev-" in speaker_dir:
+        continue
+    speaker_id = os.path.basename(speaker_dir)
+    enroll_speaker_dir = os.path.join(enroll_path, speaker_id)
+    if not os.path.exists(enroll_speaker_dir):
+        os.makedirs(enroll_speaker_dir, exist_ok=True)
+    # Check whether each speaker in librispeech_dir exists in enroll_path
+    if len(glob.glob(f"{enroll_speaker_dir}/*.wav")) == 0:
+        # Copy an audio file from librispeech_dir to enroll_speaker_dir
+        flac_files = glob.glob(f"{speaker_dir}/*/*.flac")
+        try:
+            flac_files = random.sample(flac_files, 5)
+        except:
+            print()
+            print(flac_files)
+            print(speaker_dir)
+        for flac_file in flac_files:
+            new_flac_file = os.path.join(enroll_speaker_dir, os.path.basename(flac_file))
+            # print(new_flac_file)
+            os.system(f"cp {flac_file} {new_flac_file}")
+            # Record the name of the flac_file
+            new_files.append(flac_file)
+    else:
+        print(glob.glob(f"{enroll_speaker_dir}/*.wav"))
+    
+    # Check whether the voice naming of the speaker in enroll_path is {speaker_id}.wav
+    enroll_wav_files = glob.glob(f"{enroll_speaker_dir}/*")
+    for i, enroll_wav_file in enumerate(enroll_wav_files):
+        wav_file = os.path.join(enroll_speaker_dir, speaker_id + f"_{i}.wav")
+        os.system(f"ffmpeg -i {enroll_wav_file}  -t 3  {wav_file}")
+        os.system(f"rm -rf {enroll_wav_file}")
+
+# Record the original name of the enroll wav
+with open(f"{enroll_path}/enrolled_wavs.txt", "w") as f:
+    f.write("\n".join(new_files))
\ No newline at end of file
diff --git a/data/Whisper-Sidecar-data-metadata/stat.py b/data/Whisper-Sidecar-data-metadata/stat.py
new file mode 100644
index 0000000000000000000000000000000000000000..1465dd38c1b5d2bd0e4ceae15f31ef845edcb2de
--- /dev/null
+++ b/data/Whisper-Sidecar-data-metadata/stat.py
@@ -0,0 +1,56 @@
+import glob
+
+tsv_files = glob.glob("/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/*.tsv")
+head = "\t".join(["id", "audio", "n_frames", "prompt", "tgt_text", "codec", "with_speech", "language", "speakers", "genders"])
+
+tsv_files = [
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_train.tsv",
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train.tsv",
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_train_targetLingual.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_train_targetLingual.tsv",
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_train_fixed.tsv",
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_train_fixed.tsv",
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_train_fixed.tsv",
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_train_fixed.tsv",
+    "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_train_fixed.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_test.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-2mix_test_targetLingual.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-3mix_test_targetLingual.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_test_fixed.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech2mix_test_fixed.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech3mix_test_fixed.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri2mix_test_fixed.tsv",
+    # "/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/libri3mix_test_fixed.tsv",
+]
+
+all_count = 0.0
+for tsv in tsv_files:
+    if "test" in tsv:
+        continue
+    with open(tsv, "r") as f:
+        lines = f.readline()
+        lines = f.readlines()
+    print(tsv)
+    print(len(lines))
+    count = 0.0
+    max_sec = 0
+    new_lines = []
+    for line in lines:
+        new_line = line.strip().split("\t")
+        # if float(new_line[2])/16000 > 40:
+        #     continue
+        new_lines.append("\t".join(new_line))
+        # max_sec = max_sec if max_sec > float(new_line[2])/16000 else float(new_line[2])/16000
+        count += float(new_line[2])/16000
+    # with open(tsv, "w") as f:
+    #     f.write(head + "\n" + "\n".join(new_lines)) 
+
+    print(count/60/60)
+    print(max_sec)
+    print()
+    all_count += count/60/60
+
+print(all_count)
\ No newline at end of file
diff --git a/data/de-en-mix/README.md b/data/de-en-mix/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..ca891fb064be75ad3dd329a5ddc4c8d109489254
--- /dev/null
+++ b/data/de-en-mix/README.md
@@ -0,0 +1,5 @@
+de is from commonvoice, `/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_train.tsv`
+
+en is from librispeech, `/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_train.tsv`
+
+only kept 1350 test samples.
\ No newline at end of file
diff --git a/data/de-en-mix/de-en-2mix_test.tar.gz b/data/de-en-mix/de-en-2mix_test.tar.gz
new file mode 100644
index 0000000000000000000000000000000000000000..94ac841220cf090ccfb3f4a902bd5c47dce36b5b
--- /dev/null
+++ b/data/de-en-mix/de-en-2mix_test.tar.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:821f2704b552598bf8add0f62675a65e4738782a2d5d73b17248fe845cd6612f
+size 319824887
diff --git a/data/de-en-mix/de-en-3mix_test.tar.gz b/data/de-en-mix/de-en-3mix_test.tar.gz
new file mode 100644
index 0000000000000000000000000000000000000000..5ffc0198718565a8db72094be1b0c237e8f42592
--- /dev/null
+++ b/data/de-en-mix/de-en-3mix_test.tar.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1c6a49566217d47b00f084a97330315077765d623fdc79f639d2472409092ab
+size 452491757
diff --git a/data/de-en-mix/de_test.tar.gz b/data/de-en-mix/de_test.tar.gz
new file mode 100644
index 0000000000000000000000000000000000000000..f9cc76b36cd34f0874ff46c8bfed508e865746f4
--- /dev/null
+++ b/data/de-en-mix/de_test.tar.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2173a5c71ddb6ed276cf5f4a46e7ffb5a1bf531c2c277de62861fdd9e73ed680
+size 179623462
diff --git a/data/de-en-mix/generate_metadate.py b/data/de-en-mix/generate_metadate.py
new file mode 100644
index 0000000000000000000000000000000000000000..2abae4824e00c39515fd7acb87e2c1347f8f6764
--- /dev/null
+++ b/data/de-en-mix/generate_metadate.py
@@ -0,0 +1,92 @@
+import os
+import json
+import random
+import soundfile as sf
+from tqdm import tqdm
+
+# # # de-en-2mix
+# subset = "test"
+# tsv_path = f"/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_{subset}.tsv"
+# tsv_librispeech_path = f"/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_{subset}.tsv"
+# output_jsonl_path = f"/valleblob/v-lingmeng/speech/data/de-en-mix/metadata/de-en-2mix_{subset}.jsonl"
+
+# with open(tsv_path, "r") as f1, open(tsv_librispeech_path, "r") as f2:
+#     lines = f1.readline()
+#     lines = f1.readlines()
+#     ls_lines = f2.readline()
+#     ls_lines = f2.readlines()
+
+# new_lines = []
+# count = 0
+# for i, line in enumerate(lines):
+#     new_line = {}
+#     ls_line = random.choice(ls_lines).strip().split("\t")
+#     group = [line.strip().split("\t"), ls_line]
+#     idx = random.randint(0, 1)
+#     new_line["mixed_wav"] = f'/valleblob/v-lingmeng/speech/data/de-en-mix/de-en-2mix_{subset}/{group[idx][0].split(".")[0].replace("_", "-")}_{group[1-idx][0].split(".")[0].replace("_", "-")}.wav'
+#     new_line["wavs"] = [group[idx][1], group[1-idx][1]]
+#     wav_path = group[idx][1]
+#     wav, _ = sf.read(wav_path)
+#     assert _ == 16000
+#     delay = round(random.uniform(0, int(len(wav)/16000)), 3)
+#     new_line["delays"] = [0.0, delay]
+#     # print(str(new_line))
+#     new_lines.append(json.dumps(new_line))
+#     # count += 1
+#     # if count == 3000:
+#     #     break
+    
+
+# with open(output_jsonl_path, "w") as f:
+#     f.write("\n".join(new_lines))
+    
+    
+##################
+    
+
+
+# de-en-3mix
+subset = "train"
+tsv_path = f"/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_{subset}.tsv"
+tsv_librispeech_path = f"/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_{subset}.tsv"
+output_jsonl_path = f"/valleblob/v-lingmeng/speech/data/de-en-mix/metadata/de-en-3mix_{subset}.jsonl"
+
+with open(tsv_path, "r") as f1, open(tsv_librispeech_path, "r") as f2:
+    lines = f1.readline()
+    lines = f1.readlines()
+    ls_lines = f2.readline()
+    ls_lines = f2.readlines()
+
+new_lines = []
+count = 0
+for i, line in tqdm(enumerate(lines)):
+    new_line = {}
+    ls_line_ = random.sample(ls_lines, 2)
+    group = [line.strip().split("\t"), ls_line_[0].strip().split("\t"), ls_line_[1].strip().split("\t")]
+    random.shuffle(group)
+
+    new_line["mixed_wav"] = f'/valleblob/v-lingmeng/speech/data/de-en-mix/de-en-3mix_{subset}/{group[0][0].split(".")[0].replace("_", "-")}_{group[1][0].split(".")[0].replace("_", "-")}_{group[2][0].split(".")[0].replace("_", "-")}.wav'
+    new_line["wavs"] = [group[0][1], group[1][1], group[2][1]]
+    wav_path = group[0][1]
+    wav, _ = sf.read(wav_path)
+    wav1_len = int(len(wav)/16000)
+    assert _ == 16000
+    delay1 = round(random.uniform(0, wav1_len), 3)
+    
+    wav_path = group[1][1]
+    wav, _ = sf.read(wav_path)
+    assert _ == 16000
+    wav2_len = int(len(wav)/16000)
+    max_wav_len = max(wav2_len+delay1, wav1_len)
+    delay2 = round(random.uniform(delay1+0.5, max_wav_len), 3)
+    
+    new_line["delays"] = [0.0, delay1, delay2]
+    # print(str(new_line))
+    new_lines.append(json.dumps(new_line))
+    # count += 1
+    # if count == 3000:
+    #     break
+    
+
+with open(output_jsonl_path, "w") as f:
+    f.write("\n".join(new_lines))
\ No newline at end of file
diff --git a/data/de-en-mix/generate_tsv.py b/data/de-en-mix/generate_tsv.py
new file mode 100644
index 0000000000000000000000000000000000000000..7a38949fbe4440debb3c02ff23f385db9b876cc3
--- /dev/null
+++ b/data/de-en-mix/generate_tsv.py
@@ -0,0 +1,51 @@
+import json
+import soundfile as sf
+import pandas as pd
+from tqdm import tqdm
+
+n_spks = 3
+subset = "train"
+jsonl_file = f"/valleblob/v-lingmeng/speech/data/de-en-mix/metadata/de-en-{n_spks}mix_{subset}.jsonl"
+en_df = pd.read_csv(f"/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/librispeech_{subset}.tsv", delimiter='\t', encoding='utf-8')
+de_df = pd.read_csv(f"/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de_{subset}.tsv", delimiter='\t', encoding='utf-8')
+output_tsv = f"/valleblob/v-lingmeng/speech/data/Whisper-Sidecar-data-metadata/data_for_wavllm/de-en-{n_spks}mix_{subset}.tsv"
+
+head = "\t".join(["id", "audio", "n_frames", "prompt", "tgt_text", "codec", "with_speech", "language", "speakers", "genders"])
+
+with open(jsonl_file, "r") as f:
+    lines = f.readlines()
+
+new_lines = []
+for line in tqdm(lines):
+    json_dict = json.loads(line)
+
+    id_ = json_dict['mixed_wav'].split("/")[-1].split(".wav")[0]
+    audio = json_dict['mixed_wav']
+    n_frames = str(len(sf.read(audio)[0]))
+    prompt = '"Transcribe the given audio into text, capturing the utterances of multiple speakers in English or German. List the utterances in the order of their start times, separated by "<sc>".'
+    source_ids = id_.split("_")
+    # print(de_df["id"].apply(lambda x: x.split(".")[0].replace("-","_")) )
+    # print(source_ids[1].split(".")[0])
+    # print(de_df[de_df["id"].apply(lambda x: x.split(".")[0].replace("_","-")) == source_ids[1].split(".")[0]]["tgt_text"])
+
+    tgt_texts = [
+        de_df[de_df["id"] == source_id.replace("-", "_")]["tgt_text"].iloc[0] if "common" in source_id else en_df[en_df["id"] == source_id+".flac"]["tgt_text"].iloc[0]
+        for source_id in source_ids
+    ]
+    tgt_text = " <sc> ".join(tgt_texts)
+    codec = "None"
+    with_speech = "True"
+    language = "|".join(["de" if "common" in source_id else "en" for source_id in source_ids ])
+    speakers = [
+        de_df[de_df["id"] == source_id.replace("-", "_")]["speakers"].iloc[0] if "common" in source_id else en_df[en_df["id"] == source_id+".flac"]["speakers"].iloc[0]
+        for source_id in source_ids
+    ]
+    speakers = "|".join([str(s) for s in speakers])
+    genders = "None"
+    
+    new_line = "\t".join([id_, audio, n_frames, prompt, tgt_text, codec, with_speech, language, speakers, genders])
+    new_lines.append(new_line)
+
+with open(output_tsv, "w") as f:
+    new_lines.insert(0, head)
+    f.write("\n".join(new_lines))
diff --git a/data/de-en-mix/generate_wav_from_meta_jsonl.py b/data/de-en-mix/generate_wav_from_meta_jsonl.py
new file mode 100644
index 0000000000000000000000000000000000000000..7c676e81b0c4fe795cd9c469e127c96ca56078cf
--- /dev/null
+++ b/data/de-en-mix/generate_wav_from_meta_jsonl.py
@@ -0,0 +1,60 @@
+import json
+import os
+import glob
+import soundfile
+import librosa
+from tqdm import tqdm
+import numpy as np
+
+def get_delayed_audio(wav_file, delay, sampling_rate=16000):
+    audio, _ = soundfile.read(wav_file)
+    delay_frame = int(delay * sampling_rate)
+    if delay_frame != 0:
+        audio = np.append(np.zeros(delay_frame), audio)
+    return audio
+
+def mix_audio(wav_files, delays, weight):
+    for i, wav_file in enumerate(wav_files):
+        if i == 0:
+            audio = get_delayed_audio(wav_file, delays[i])
+            audio = weight[i] * audio
+            
+        else:
+            additional_audio = get_delayed_audio(wav_file, delays[i])
+            additional_audio = weight[i] * additional_audio
+            # tune length & sum up to audio
+            target_length = max(len(audio), len(additional_audio))
+            # print(additional_audio.shape)
+            audio = librosa.util.fix_length(audio, size=target_length)
+            additional_audio = librosa.util.fix_length(additional_audio, size=target_length)
+            audio = audio + additional_audio
+    return audio
+
+
+jsonl_path = "/valleblob/v-lingmeng/speech/data/de-en-mix/metadata/de-en-3mix_train.jsonl"
+librispeech_dir = f""
+output_dir = ""
+
+
+with open(jsonl_path, 'r', encoding='utf-8') as file:
+    json_list = [json.loads(line.strip()) for line in file]
+
+for line in tqdm(json_list):
+    audio_name = output_dir +  line['mixed_wav']
+    _audio_name = audio_name.split("/")[-1].split("_")
+    weight = [1.0 if "common" in _name else 0.3 for _name in _audio_name]   # 德语声音非常小
+
+    if not os.path.exists(os.path.split(audio_name)[0]):
+        os.makedirs(os.path.split(audio_name)[0])
+    
+    source_ids = line['wavs']
+    source_files = [librispeech_dir + i.replace(".wav", ".flac") for i in source_ids]
+    delays = line["delays"]
+    # ends = [l["end"] for l in line['sentences']]
+    
+    # for source_audio, start, end in zip(source_auidos, starts, ends):
+    mixed_audio = mix_audio(source_files, delays, weight)
+    soundfile.write(audio_name, mixed_audio, samplerate=16000)
+    # print(audio_name)
+    
+# print(delays)
diff --git a/data/de-en-mix/metadata/de-en-2mix_test.jsonl b/data/de-en-mix/metadata/de-en-2mix_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..ef4192a77333813df22bd6d918be70fa56a548d8
--- /dev/null
+++ b/data/de-en-mix/metadata/de-en-2mix_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dfc774785b5a6f4a886c993b9150c6b43cd4fdc8f578db0b9fe2adb832aa08a
+size 4911835
diff --git a/data/de-en-mix/metadata/de-en-2mix_train.jsonl b/data/de-en-mix/metadata/de-en-2mix_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..55635d9602120feec6529bf719765ada00cb7912
--- /dev/null
+++ b/data/de-en-mix/metadata/de-en-2mix_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51190146b9cdcd9f450f4644f619f972ea498e74a672733e14d14a809f1abe12
+size 47206360
diff --git a/data/de-en-mix/metadata/de-en-3mix_test.jsonl b/data/de-en-mix/metadata/de-en-3mix_test.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e05fbedf3420fbfe6bed0f0622d17bd581237563
--- /dev/null
+++ b/data/de-en-mix/metadata/de-en-3mix_test.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a2e30c4cc607151c514d4716fcdf1da5c8ee51deb79548d4a4d7e0fb81126ee
+size 6637883
diff --git a/data/de-en-mix/metadata/de-en-3mix_train.jsonl b/data/de-en-mix/metadata/de-en-3mix_train.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..3be404e1f39bd0861a1cd6cdad3435084f18b19b
--- /dev/null
+++ b/data/de-en-mix/metadata/de-en-3mix_train.jsonl
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7fbad74fc6a0ca52783b0498d1556a57e0bc3162ae28028485ff4096ee5b6b6
+size 64220827
diff --git a/data/reference_enroll_audio.tar.gz b/data/reference_enroll_audio.tar.gz
new file mode 100644
index 0000000000000000000000000000000000000000..c0114e9ff8e555bb1ce2c7191bfd641a3be60927
--- /dev/null
+++ b/data/reference_enroll_audio.tar.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62a481aa7bcbc9eb8c27c48ac8d1a8f69d2bd1f6f41558547d4a3511a25de203
+size 968316801
diff --git a/llama_model/llama-2-7b-chat/checklist.chk b/llama_model/llama-2-7b-chat/checklist.chk
new file mode 100644
index 0000000000000000000000000000000000000000..150da5509a7461cf6c16a6e1acb0ca6329249b66
--- /dev/null
+++ b/llama_model/llama-2-7b-chat/checklist.chk
@@ -0,0 +1,2 @@
+0c4837f3ef97f648452f91faed308a07  consolidated.00.pth
+1c39bc3c6b51079fd807cc105b86c9df  params.json
diff --git a/llama_model/llama-2-7b-chat/consolidated.00.pth b/llama_model/llama-2-7b-chat/consolidated.00.pth
new file mode 100644
index 0000000000000000000000000000000000000000..a1e414306a4aa4a7d54985c55a655307299da8c4
--- /dev/null
+++ b/llama_model/llama-2-7b-chat/consolidated.00.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6234f92a9191a4887b65a7f14a9692b4af3beffa2a26359869daf36bdf71b8d8
+size 13476925163
diff --git a/llama_model/llama-2-7b-chat/params.json b/llama_model/llama-2-7b-chat/params.json
new file mode 100644
index 0000000000000000000000000000000000000000..e4cef7be9cf058c8a44bb166ea0adbcf94173f97
--- /dev/null
+++ b/llama_model/llama-2-7b-chat/params.json
@@ -0,0 +1 @@
+{"dim": 4096, "multiple_of": 256, "n_heads": 32, "n_layers": 32, "norm_eps": 1e-06, "vocab_size": -1}
diff --git a/llama_model/llama/tokenizer.model b/llama_model/llama/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899
--- /dev/null
+++ b/llama_model/llama/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/llama_model/llama/tokenizer_checklist.chk b/llama_model/llama/tokenizer_checklist.chk
new file mode 100644
index 0000000000000000000000000000000000000000..4531f05cde0f2f2cb2d44055cf08e1d467d40196
--- /dev/null
+++ b/llama_model/llama/tokenizer_checklist.chk
@@ -0,0 +1 @@
+eeec4125e9c7560836b4873b6f8e3025  tokenizer.model