成功训练

examples/asr/asr_eou/speech_to_text_rnnt_eou_train_number.py

@@ -3,70 +3,6 @@
     # huggingface_echodict\asr_rnnt_eou_from_scratch\papers\arXiv-1211.3711v1\training_number.py
     # 训练中文数字识别
 
-"""
-Example usage:
-
-1. Prepare dataset based on <NeMo Root>/nemo/collections/asr/data/audio_to_eou_label_lhotse.py
-  Specifically, each sample in the jsonl manifest should have the following fields:
-  {
-    "audio_filepath": "/path/to/audio.wav",
-    "text": "The text of the audio."
-    "offset": 0.0,  # offset of the audio, in seconds
-    "duration": 3.0,  # duration of the audio, in seconds
-    "sou_time": 0.2,  # start of utterance time, in seconds
-    "eou_time": 1.5,  # end of utterance time, in seconds
-  }
-
-2. If using a normal ASR model as initialization:
-    -  Add special tokens <EOU> and <EOB> to the tokenizer of pretrained model, by refering to the script
-        <NeMo Root>/scripts/asr_eou/tokenizers/add_special_tokens_to_sentencepiece.py
-    - If pretrained model is HybridRNNTCTCBPEModel, convert it to RNNT using the script
-        <NeMo Root>/examples/asr/asr_hybrid_transducer_ctc/helpers/convert_nemo_asr_hybrid_to_ctc.py
-
-3. Run the following command to train the ASR-EOU model:
-```bash
-#!/bin/bash
-
-TRAIN_MANIFEST=/path/to/train_manifest.json
-VAL_MANIFEST=/path/to/val_manifest.json
-NOISE_MANIFEST=/path/to/noise_manifest.json
-
-PRETRAINED_NEMO=/path/to/pretrained_model.nemo
-TOKENIZER_DIR=/path/to/tokenizer_dir
-
-BATCH_SIZE=16
-NUM_WORKERS=8
-LIMIT_TRAIN_BATCHES=1000
-VAL_CHECK_INTERVAL=1000
-MAX_STEPS=1000000
-
-EXP_NAME=fastconformer_transducer_bpe_streaming_eou
-SCRIPT=${NEMO_PATH}/examples/asr/asr_eou/speech_to_text_rnnt_eou_train.py
-CONFIG_PATH=${NEMO_PATH}/examples/asr/conf/asr_eou
-CONFIG_NAME=fastconformer_transducer_bpe_streaming
-
-CUDA_VISIBLE_DEVICES=0 python $SCRIPT \
-    --config-path $CONFIG_PATH \
-    --config-name $CONFIG_NAME \
-    ++init_from_nemo_model=$PRETRAINED_NEMO \
-    model.encoder.att_context_size="[70,1]" \
-    model.tokenizer.dir=$TOKENIZER_DIR \
-    model.train_ds.manifest_filepath=$TRAIN_MANIFEST \
-    model.train_ds.augmentor.noise.manifest_path=$NOISE_MANIFEST \
-    model.validation_ds.manifest_filepath=$VAL_MANIFEST \
-    model.train_ds.batch_size=$BATCH_SIZE \
-    model.train_ds.num_workers=$NUM_WORKERS \
-    model.validation_ds.batch_size=$BATCH_SIZE \
-    model.validation_ds.num_workers=$NUM_WORKERS \
-    ~model.test_ds \
-    trainer.limit_train_batches=$LIMIT_TRAIN_BATCHES \
-    trainer.val_check_interval=$VAL_CHECK_INTERVAL \
-    trainer.max_steps=$MAX_STEPS \
-    exp_manager.name=$EXP_NAME
-```
-
-"""
-
 import os
 import sys
 sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '../../..')))
@@ -343,6 +279,81 @@ def main(cfg):
 
 if __name__ == '__main__':
     import sys
+    import os
+    import json
+    import wave
+    import subprocess
+    import shutil
+
+    txt_path = r"e:\huggingface_echodict\NeMo\data\tts_dataset\labels.txt"
+    manifest_path = txt_path.replace(".txt", "_manifest.jsonl")
+    text_corpus_path = txt_path.replace(".txt", "_text.txt")
+    tokenizer_dir = os.path.join(os.path.dirname(txt_path), "tokenizer_eou")
+    
+    # 动态生成 manifest 和文本语料
+    if not os.path.exists(manifest_path) or not os.path.exists(text_corpus_path):
+        base_dir = os.path.dirname(txt_path)
+        with open(txt_path, 'r', encoding='utf-8') as f_in, \
+             open(manifest_path, 'w', encoding='utf-8') as f_out, \
+             open(text_corpus_path, 'w', encoding='utf-8') as f_txt:
+            for line in f_in:
+                line = line.strip()
+                if not line:
+                    continue
+                parts = line.split('\t')
+                if len(parts) == 2:
+                    audio_file, text = parts
+                    audio_filepath = os.path.join(base_dir, audio_file)
+                    if not os.path.exists(audio_filepath):
+                        print(f"Warning: {audio_filepath} not found.")
+                        continue
+                    try:
+                        with wave.open(audio_filepath, 'r') as w:
+                            frames = w.getnframes()
+                            rate = w.getframerate()
+                            duration = frames / float(rate)
+                    except Exception as e:
+                        print(f"Error reading {audio_filepath}: {e}")
+                        continue
+                    
+                    item = {
+                        "audio_filepath": audio_filepath,
+                        "duration": duration,
+                        "text": text,
+                    }
+                    f_out.write(json.dumps(item, ensure_ascii=False) + '\n')
+                    f_txt.write(text + '\n')
+
+    # 生成 Tokenizer 并添加 EOU/EOB tokens
+    if not os.path.exists(os.path.join(tokenizer_dir, "tokenizer.model")):
+        print("Generating tokenizer...")
+        from nemo.collections.common.tokenizers.sentencepiece_tokenizer import create_spt_model
+        
+        # 1. 训练基础 Tokenizer
+        temp_tokenizer_dir = tokenizer_dir + "_temp"
+        os.makedirs(temp_tokenizer_dir, exist_ok=True)
+        create_spt_model(
+            data_file=text_corpus_path,
+            vocab_size=32,
+            sample_size=-1,
+            do_lower_case=True,
+            output_dir=temp_tokenizer_dir,
+            tokenizer_type="bpe",
+            character_coverage=1.0,
+        )
+        
+        # 2. 注入 EOU/EOB tokens
+        # 使用 NeMo 提供的工具脚本
+        add_special_tokens_script = os.path.abspath(os.path.join(os.path.dirname(__file__), "../../../scripts/asr_eou/tokenizers/add_special_tokens_to_sentencepiece.py"))
+        base_model_path = os.path.join(temp_tokenizer_dir, "tokenizer.model")
+        
+        subprocess.check_call([
+            sys.executable, add_special_tokens_script,
+            "--input_file", base_model_path,
+            "--output_dir", tokenizer_dir
+        ])
+        print(f"Tokenizer generated successfully at {tokenizer_dir}")
+
     sys.argv.extend([
         '--config-path', '../conf/asr_eou/',
         '--config-name', 'fastconformer_transducer_bpe_streaming_large',
@@ -353,15 +364,11 @@ if __name__ == '__main__':
         'exp_manager.checkpoint_callback_params.save_top_k=1',
         '++trainer.check_val_every_n_epoch=1',
         '++model.encoder.conv_norm_type=layer_norm',
-        'model.tokenizer.dir=data/common_voice_11_0/ja/tokenizers/tokenizer_spe_bpe_v2491_eou',
-        'model.train_ds.tarred_audio_filepaths=data/common_voice_11_0/ja/train_tarred_1bk/audio__OP_0..1023_CL_.tar',
-        '++model.train_ds.is_tarred=true',
-        '++model.train_ds.tarred_dataset_resolve_paths=false',
-        '++model.train_ds.is_tarred_audio=true',
-        'model.train_ds.manifest_filepath=data/common_voice_11_0/ja/train_tarred_1bk/tarred_audio_manifest.json',
+        f'model.tokenizer.dir={tokenizer_dir}',
+        f'model.train_ds.manifest_filepath={manifest_path}',
         '~model.train_ds.augmentor.noise',
-        'model.validation_ds.manifest_filepath=data/common_voice_11_0/ja/validation_tarred_1bk/tarred_audio_manifest.json',
-        'model.test_ds.manifest_filepath=data/common_voice_11_0/ja/test_tarred_1bk/tarred_audio_manifest.json',
+        f'model.validation_ds.manifest_filepath={manifest_path}',
+        f'model.test_ds.manifest_filepath={manifest_path}',
         'trainer.max_epochs=1',
         'trainer.devices=1',
         'trainer.accelerator=gpu',

scripts/asr_eou/tokenizers/add_special_tokens_to_sentencepiece.py

@@ -14,6 +14,9 @@
 
 import os
 
+import sys
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '../../..')))
+
 os.environ["PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION"] = "python"
 import logging
 import sys