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
perf: optimize _reset_runtime_configs to skip deepcopy of immutable objects
游雁4bb33425a865
diff --git a/funasr/auto/auto_model.py b/funasr/auto/auto_model.pyindex dc7550c30..9b0054e32 100644--- a/funasr/auto/auto_model.py+++ b/funasr/auto/auto_model.py@@ -801,6 +801,10 @@ class AutoModel:baseline["kwargs"] = copy.deepcopy(self.kwargs)self._base_kwargs_map = baseline+ _IMMUTABLE_KWARGS_KEYS = frozenset([+ "token_list", "tokenizer", "frontend", "model", "init_param", "model_path",+ ])+def _reset_runtime_configs(self):"""Ensure runtime kwargs reset to baseline defaults before inference."""base_map = getattr(self, "_base_kwargs_map", None)@@ -808,7 +812,12 @@ class AutoModel:returnfor name, base in base_map.items():- restored = copy.deepcopy(base)+ restored = {}+ for k, v in base.items():+ if k in self._IMMUTABLE_KWARGS_KEYS or not isinstance(v, (dict, list)):+ restored[k] = v+ else:+ restored[k] = copy.deepcopy(v)setattr(self, name, restored)ncpu = _resolve_ncpu(self.kwargs, 4)diff --git a/tests_models/test_paraformer_streaming.py b/tests_models/test_paraformer_streaming.pyindex 99aec136e..5846de7c8 100644--- a/tests_models/test_paraformer_streaming.py+++ b/tests_models/test_paraformer_streaming.py@@ -1,9 +1,11 @@#!/usr/bin/env python3-"""Test Paraformer-Streaming: online/streaming ASR"""+"""Test Paraformer-Streaming: chunk-by-chunk streaming ASR"""import sysimport time+import osdef main():+ import soundfilefrom funasr import AutoModelprint("[Paraformer-Streaming] Loading model...")@@ -12,21 +14,49 @@ def main():model="iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online",device="cpu",disable_update=True,+ disable_pbar=True,)- print("[Paraformer-Streaming] Model loaded in %.1fs" % (time.time()-t0))+ print("[Paraformer-Streaming] Model loaded in %.1fs" % (time.time() - t0))- print("[Paraformer-Streaming] Running inference...")+ wav_file = os.path.join(model.model_path, "example/asr_example.wav")+ speech, sample_rate = soundfile.read(wav_file)++ chunk_size = [0, 10, 5]+ encoder_chunk_look_back = 4+ decoder_chunk_look_back = 1+ chunk_stride = chunk_size[1] * 960 # 600ms++ print("[Paraformer-Streaming] Running streaming inference (%.2fs audio, %d chunks)..." % (+ len(speech) / sample_rate, int((len(speech) - 1) / chunk_stride + 1)))t0 = time.time()- test_wav = "https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav"- res = model.generate(input=test_wav, cache={}, is_final=True)- print("[Paraformer-Streaming] Inference done in %.1fs" % (time.time()-t0))- print("[Paraformer-Streaming] Result: %s" % res)- if res and len(res) > 0 and "text" in res[0]:+ cache = {}+ total_chunk_num = int((len(speech) - 1) / chunk_stride + 1)+ all_text = ""++ for i in range(total_chunk_num):+ speech_chunk = speech[i * chunk_stride:(i + 1) * chunk_stride]+ is_final = i == total_chunk_num - 1+ res = model.generate(+ input=speech_chunk,+ cache=cache,+ is_final=is_final,+ chunk_size=chunk_size,+ encoder_chunk_look_back=encoder_chunk_look_back,+ decoder_chunk_look_back=decoder_chunk_look_back,+ )+ txt = res[0].get("text", "") if res else ""+ all_text += txt++ print("[Paraformer-Streaming] Inference done in %.1fs" % (time.time() - t0))+ print("[Paraformer-Streaming] Result: '%s'" % all_text)++ expected = "欢迎大家来体验达摩院推出的语音识别模型"+ if expected in all_text:print("[Paraformer-Streaming] PASSED")return 0else:- print("[Paraformer-Streaming] FAILED - no text in result")+ print("[Paraformer-Streaming] FAILED - expected text not found")return 1if __name__ == "__main__":