Sherpa-CTC (ru)

.gitattributes

@@ -59,3 +59,4 @@ ja/sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8/test_wavs/test_ja_1.wav
 ja/sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8/test_wavs/test_ja_2.wav filter=lfs diff=lfs merge=lfs -text
 ru/sherpa-onnx-nemo-ctc-giga-am-russian-2024-10-24/test_wavs/example.wav filter=lfs diff=lfs merge=lfs -text
 ru/sherpa-onnx-nemo-ctc-giga-am-russian-2024-10-24/test_wavs/long_example.wav filter=lfs diff=lfs merge=lfs -text
+ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/test_wavs/example.wav filter=lfs diff=lfs merge=lfs -text

ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz 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
+*.wav filter=lfs diff=lfs merge=lfs -text

ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/export-onnx-ctc-v2.py

@@ -0,0 +1,64 @@
+#!/usr/bin/env python3
+import gigaam
+import onnx
+import torch
+from onnxruntime.quantization import QuantType, quantize_dynamic
+
+
+def add_meta_data(filename: str, meta_data: dict[str, str]):
+    """Add meta data to an ONNX model. It is changed in-place.
+
+    Args:
+      filename:
+        Filename of the ONNX model to be changed.
+      meta_data:
+        Key-value pairs.
+    """
+    model = onnx.load(filename)
+    while len(model.metadata_props):
+        model.metadata_props.pop()
+
+    for key, value in meta_data.items():
+        meta = model.metadata_props.add()
+        meta.key = key
+        meta.value = str(value)
+
+    onnx.save(model, filename)
+
+
+def main() -> None:
+    model_name = "v2_ctc"
+    model = gigaam.load_model(
+        model_name, fp16_encoder=False, use_flash=False, download_root="."
+    )
+
+    # use characters
+    # space is 0
+    # <blk> is the last token
+    with open("./tokens.txt", "w", encoding="utf-8") as f:
+        for i, s in enumerate(model.cfg["labels"]):
+            f.write(f"{s} {i}\n")
+        f.write(f"<blk> {i+1}\n")
+        print("Saved to tokens.txt")
+    model.to_onnx(".")
+    meta_data = {
+        "vocab_size": len(model.cfg["labels"]) + 1,
+        "normalize_type": "",
+        "subsampling_factor": 4,
+        "model_type": "EncDecCTCModel",
+        "version": "1",
+        "model_author": "https://github.com/salute-developers/GigaAM",
+        "license": "https://github.com/salute-developers/GigaAM/blob/main/LICENSE",
+        "language": "Russian",
+        "is_giga_am": 1,
+    }
+    add_meta_data(f"./{model_name}.onnx", meta_data)
+    quantize_dynamic(
+        model_input=f"./{model_name}.onnx",
+        model_output="./model.int8.onnx",
+        weight_type=QuantType.QUInt8,
+    )
+
+
+if __name__ == "__main__":
+    main()

ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/model.int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0ce4aef25f58d495781ee8f05320d9e51b821f47804e07aa6549b53a72f67e8
+size 236457977

ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/run-ctc-v2.sh

@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+
+set -ex
+
+function install_gigaam() {
+  curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
+  python3 get-pip.py
+  pip install torch==2.4.0 torchaudio==2.4.0 -f https://download.pytorch.org/whl/torch_stable.html
+  pip install -qq wget text-unidecode "matplotlib>=3.3.2" onnx onnxruntime==1.17.1 pybind11 Cython einops kaldi-native-fbank soundfile librosa
+
+  BRANCH='main'
+  python3 -m pip install git+https://github.com/salute-developers/GigaAM.git@$BRANCH#egg=gigaam
+
+  python3 -m pip install -qq kaldi-native-fbank
+  pip install numpy==1.26.4
+}
+
+function download_files() {
+  curl -SL -O https://huggingface.co/csukuangfj/tmp-files/resolve/main/GigaAM/example.wav
+  curl -SL -O https://github.com/salute-developers/GigaAM/blob/main/LICENSE
+}
+
+install_gigaam
+download_files
+
+python3 ./export-onnx-ctc-v2.py
+ls -lh
+python3 ./test-onnx-ctc.py

ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/test-onnx-ctc.py

@@ -0,0 +1,155 @@
+#!/usr/bin/env python3
+# Copyright      2024  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+# https://github.com/salute-developers/GigaAM
+
+import kaldi_native_fbank as knf
+import librosa
+import numpy as np
+import onnxruntime as ort
+import soundfile as sf
+import torch
+
+
+def create_fbank():
+    opts = knf.FbankOptions()
+    opts.frame_opts.dither = 0
+    opts.frame_opts.remove_dc_offset = False
+    opts.frame_opts.preemph_coeff = 0
+    opts.frame_opts.window_type = "hann"
+
+    opts.frame_opts.round_to_power_of_two = False
+
+    opts.mel_opts.low_freq = 0
+    opts.mel_opts.high_freq = 8000
+    opts.mel_opts.num_bins = 64
+
+    fbank = knf.OnlineFbank(opts)
+    return fbank
+
+
+def compute_features(audio, fbank) -> np.ndarray:
+    """
+    Args:
+      audio: (num_samples,), np.float32
+      fbank: the fbank extractor
+    Returns:
+      features: (num_frames, feat_dim), np.float32
+    """
+    assert len(audio.shape) == 1, audio.shape
+    fbank.accept_waveform(16000, audio)
+    ans = []
+    processed = 0
+    while processed < fbank.num_frames_ready:
+        ans.append(np.array(fbank.get_frame(processed)))
+        processed += 1
+    ans = np.stack(ans)
+    return ans
+
+
+def display(sess):
+    print("==========Input==========")
+    for i in sess.get_inputs():
+        print(i)
+    print("==========Output==========")
+    for i in sess.get_outputs():
+        print(i)
+
+
+"""
+==========Input==========
+NodeArg(name='audio_signal', type='tensor(float)', shape=['audio_signal_dynamic_axes_1', 64, 'audio_signal_dynamic_axes_2'])
+NodeArg(name='length', type='tensor(int64)', shape=['length_dynamic_axes_1'])
+==========Output==========
+NodeArg(name='logprobs', type='tensor(float)', shape=['logprobs_dynamic_axes_1', 'logprobs_dynamic_axes_2', 34])
+"""
+
+
+class OnnxModel:
+    def __init__(
+        self,
+        filename: str,
+    ):
+        session_opts = ort.SessionOptions()
+        session_opts.inter_op_num_threads = 1
+        session_opts.intra_op_num_threads = 1
+
+        self.model = ort.InferenceSession(
+            filename,
+            sess_options=session_opts,
+            providers=["CPUExecutionProvider"],
+        )
+        display(self.model)
+
+    def __call__(self, x: np.ndarray):
+        # x: (T, C)
+        x = torch.from_numpy(x)
+        x = x.t().unsqueeze(0)
+        # x: [1, C, T]
+        x_lens = torch.tensor([x.shape[-1]], dtype=torch.int64)
+
+        log_probs = self.model.run(
+            [
+                self.model.get_outputs()[0].name,
+            ],
+            {
+                self.model.get_inputs()[0].name: x.numpy(),
+                self.model.get_inputs()[1].name: x_lens.numpy(),
+            },
+        )[0]
+        # [batch_size, T, dim]
+        return log_probs
+
+
+def main():
+    filename = "./model.int8.onnx"
+    tokens = "./tokens.txt"
+    wav = "./example.wav"
+
+    model = OnnxModel(filename)
+
+    id2token = dict()
+    with open(tokens, encoding="utf-8") as f:
+        for line in f:
+            fields = line.split()
+            if len(fields) == 1:
+                id2token[int(fields[0])] = " "
+            else:
+                t, idx = fields
+                id2token[int(idx)] = t
+
+    fbank = create_fbank()
+    audio, sample_rate = sf.read(wav, dtype="float32", always_2d=True)
+    audio = audio[:, 0]  # only use the first channel
+    if sample_rate != 16000:
+        audio = librosa.resample(
+            audio,
+            orig_sr=sample_rate,
+            target_sr=16000,
+        )
+        sample_rate = 16000
+
+    features = compute_features(audio, fbank)
+    print("features.shape", features.shape)
+
+    blank = len(id2token) - 1
+    prev = -1
+    ans = []
+    log_probs = model(features)
+    print("log_probs", log_probs.shape)
+    log_probs = torch.from_numpy(log_probs)[0]
+    ids = torch.argmax(log_probs, dim=1).tolist()
+    for i in ids:
+        if i != blank and i != prev:
+            ans.append(i)
+        prev = i
+
+    tokens = [id2token[i] for i in ans]
+
+    text = "".join(tokens)
+    print(wav)
+    print(text)
+
+
+if __name__ == "__main__":
+    main()

ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/test_wavs/example.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8aaaa18a5098d7c6de0595ae7ac1e64cacd0d4022af3595213bdaf23be77e69
+size 361324

ru/sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19/tokens.txt

@@ -0,0 +1,34 @@
+  0
+а 1
+б 2
+в 3
+г 4
+д 5
+е 6
+ж 7
+з 8
+и 9
+й 10
+к 11
+л 12
+м 13
+н 14
+о 15
+п 16
+р 17
+с 18
+т 19
+у 20
+ф 21
+х 22
+ц 23
+ч 24
+ш 25
+щ 26
+ъ 27
+ы 28
+ь 29
+э 30
+ю 31
+я 32
+<blk> 33