Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
fix:添加音频码率验证以防止模型异常 (#2219)
pointerhacker37d6930561f4
diff --git a/funasr/utils/load_utils.py b/funasr/utils/load_utils.pyindex 37b3dbfbf..aabf5d9f3 100644--- a/funasr/utils/load_utils.py+++ b/funasr/utils/load_utils.py@@ -1,6 +1,7 @@import osimport torchimport json+from io import BytesIOimport torch.distributed as distimport numpy as npimport kaldiio@@ -9,6 +10,7 @@ import torchaudioimport timeimport loggingfrom torch.nn.utils.rnn import pad_sequence+from pydub import AudioSegmenttry:from funasr.download.file import download_from_url@@ -136,6 +138,7 @@ def load_audio_text_image_video(def load_bytes(input):+ input = validate_frame_rate(input)middle_data = np.frombuffer(input, dtype=np.int16)middle_data = np.asarray(middle_data)if middle_data.dtype.kind not in "iu":@@ -150,6 +153,31 @@ def load_bytes(input):array = np.frombuffer((middle_data.astype(dtype) - offset) / abs_max, dtype=np.float32)return array+def validate_frame_rate(+ input,+ fs: int = 16000,+):++ # 将文件读取为字节流+ byte_data = BytesIO(input)++ # 使用 pydub 加载音频+ audio = AudioSegment.from_file(byte_data)++ # 确保采样率为 16000 Hz+ if audio.frame_rate != fs:+ audio = audio.set_frame_rate(fs)++ # 将重新采样后的音频导出为字节流+ output = BytesIO()+ audio.export(output, format="wav")+ output.seek(0)++ # 获取重新采样后的字节流数据+ input = output.read()++ return input+def extract_fbank(data, data_len=None, data_type: str = "sound", frontend=None, **kwargs):if isinstance(data, np.ndarray):