Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

fix(onnx): load non-.wav audio paths with librosa instead of pydub

Tanishqa3b21c4ef348
diff --git a/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.py b/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.py
index d822e7318..517252cd8 100644
--- a/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.py
+++ b/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.py
@@ -181,23 +181,7 @@ class Paraformer:
plt.savefig(plotname, bbox_inches="tight")
def load_data(self, wav_content: Union[str, np.ndarray, List[str]], fs: int = None) -> List:
- def convert_to_wav(input_path, output_path):
- from pydub import AudioSegment
- try:
- audio = AudioSegment.from_mp3(input_path)
- audio.export(output_path, format="wav")
- print("音频文件为mp3格式,已转换为wav格式")
-
- except Exception as e:
- print(f"转换失败:{e}")
-
def load_wav(path: str) -> np.ndarray:
- if not path.lower().endswith('.wav'):
- import os
- input_path = path
- path = os.path.splitext(path)[0]+'.wav'
- convert_to_wav(input_path,path) #将mp3格式转换成wav格式
-
waveform, _ = librosa.load(path, sr=fs)
return waveform
diff --git a/runtime/python/onnxruntime/funasr_onnx/paraformer_online_bin.py b/runtime/python/onnxruntime/funasr_onnx/paraformer_online_bin.py
index 6c86bcd46..4d353b733 100644
--- a/runtime/python/onnxruntime/funasr_onnx/paraformer_online_bin.py
+++ b/runtime/python/onnxruntime/funasr_onnx/paraformer_online_bin.py
@@ -219,24 +219,7 @@ class Paraformer:
return asr_res
def load_data(self, wav_content: Union[str, np.ndarray, List[str]], fs: int = None) -> List:
-
- def convert_to_wav(input_path, output_path):
- from pydub import AudioSegment
- try:
- audio = AudioSegment.from_mp3(input_path)
- audio.export(output_path, format="wav")
- print("音频文件为mp3格式,已转换为wav格式")
-
- except Exception as e:
- print(f"转换失败:{e}")
-
def load_wav(path: str) -> np.ndarray:
- if not path.lower().endswith('.wav'):
- import os
- input_path = path
- path = os.path.splitext(path)[0]+'.wav'
- convert_to_wav(input_path,path) #将mp3格式转换成wav格式
-
waveform, _ = librosa.load(path, sr=fs)
return waveform
diff --git a/runtime/python/onnxruntime/funasr_onnx/sensevoice_bin.py b/runtime/python/onnxruntime/funasr_onnx/sensevoice_bin.py
index 0bc265ebb..d3630db5a 100644
--- a/runtime/python/onnxruntime/funasr_onnx/sensevoice_bin.py
+++ b/runtime/python/onnxruntime/funasr_onnx/sensevoice_bin.py
@@ -202,24 +202,7 @@ class SenseVoiceSmall:
return asr_res
def load_data(self, wav_content: Union[str, np.ndarray, List[str]], fs: int = None) -> List:
-
- def convert_to_wav(input_path, output_path):
- from pydub import AudioSegment
- try:
- audio = AudioSegment.from_mp3(input_path)
- audio.export(output_path, format="wav")
- print("音频文件为mp3格式,已转换为wav格式")
-
- except Exception as e:
- print(f"转换失败:{e}")
-
def load_wav(path: str) -> np.ndarray:
- if not path.lower().endswith('.wav'):
- import os
- input_path = path
- path = os.path.splitext(path)[0]+'.wav'
- convert_to_wav(input_path,path) #将mp3格式转换成wav格式
-
waveform, _ = librosa.load(path, sr=fs)
return waveform
diff --git a/runtime/python/onnxruntime/funasr_onnx/vad_bin.py b/runtime/python/onnxruntime/funasr_onnx/vad_bin.py
index 5d9d8fd4e..74cfa2944 100644
--- a/runtime/python/onnxruntime/funasr_onnx/vad_bin.py
+++ b/runtime/python/onnxruntime/funasr_onnx/vad_bin.py
@@ -150,24 +150,7 @@ class Fsmn_vad:
return segments
def load_data(self, wav_content: Union[str, np.ndarray, List[str]], fs: int = None) -> List:
-
- def convert_to_wav(input_path, output_path):
- from pydub import AudioSegment
- try:
- audio = AudioSegment.from_mp3(input_path)
- audio.export(output_path, format="wav")
- print("音频文件为mp3格式,已转换为wav格式")
-
- except Exception as e:
- print(f"转换失败:{e}")
-
def load_wav(path: str) -> np.ndarray:
- if not path.lower().endswith('.wav'):
- import os
- input_path = path
- path = os.path.splitext(path)[0]+'.wav'
- convert_to_wav(input_path,path) #将mp3格式转换成wav格式
-
waveform, _ = librosa.load(path, sr=fs)
return waveform
diff --git a/tests/test_runtime_eval_native.py b/tests/test_runtime_eval_native.py
index 62747fb2c..e0e615480 100644
--- a/tests/test_runtime_eval_native.py
+++ b/tests/test_runtime_eval_native.py
@@ -273,3 +273,40 @@ def test_hotword_wrappers_decode_every_batch(
assert [row["preds"][0] for row in alone] == ["你" * 4, "你" * 11, "你" * 17]
# More files than batch_size: every batch must return its results.
assert model(paths, hotwords) == alone
+
+
[email protected]("extension", ["flac", "ogg", "mp3"])
+def test_onnx_wrappers_read_non_wav_paths(native_fixture, tmp_path, monkeypatch, extension):
+ np = pytest.importorskip("numpy")
+ librosa = pytest.importorskip("librosa")
+ soundfile = pytest.importorskip("soundfile")
+ if extension.upper() not in soundfile.available_formats():
+ pytest.skip(f"libsndfile {soundfile.__libsndfile_version__} cannot write {extension}")
+ directory, _, _ = native_fixture
+ monkeypatch.syspath_prepend(str(ROOT / "runtime/python/onnxruntime"))
+ funasr_onnx = importlib.import_module("funasr_onnx")
+ online = importlib.import_module("funasr_onnx.paraformer_online_bin")
+ audio = tmp_path / f"talk.{extension}"
+ tone = np.sin(np.arange(48000) * 2 * np.pi * 440 / 16000).astype(np.float32) * 0.1
+ soundfile.write(str(audio), tone, 16000)
+ # A different recording with the same stem must not be read in its place.
+ soundfile.write(str(tmp_path / "talk.wav"), np.zeros(16000, dtype=np.float32), 16000)
+ files = {path.name: path.read_bytes() for path in tmp_path.iterdir()}
+ expected = librosa.load(str(audio), sr=16000)[0]
+ wrappers = [
+ funasr_onnx.Paraformer,
+ funasr_onnx.ContextualParaformer,
+ funasr_onnx.SeacoParaformer,
+ online.Paraformer,
+ funasr_onnx.Fsmn_vad,
+ funasr_onnx.SenseVoiceSmall,
+ ]
+ for wrapper in wrappers:
+ # load_data uses no model state, so skip __init__ and its model files.
+ waveforms = wrapper.__new__(wrapper).load_data(str(audio), 16000)
+ assert len(waveforms) == 1
+ name = f"{wrapper.__module__}.{wrapper.__name__}"
+ np.testing.assert_array_equal(waveforms[0], expected, err_msg=name)
+ model = funasr_onnx.Paraformer(str(directory), device_id=DEVICE, intra_op_num_threads=1)
+ assert model(str(audio)) == model(expected)
+ assert {path.name: path.read_bytes() for path in tmp_path.iterdir()} == files