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 bug
游雁7fc17b1aafb0
diff --git a/examples/industrial_data_pretraining/llm_asr/demo_speech2text.py b/examples/industrial_data_pretraining/llm_asr/demo_speech2text.pyindex dfbe95b35..e5e3e2302 100644--- a/examples/industrial_data_pretraining/llm_asr/demo_speech2text.py+++ b/examples/industrial_data_pretraining/llm_asr/demo_speech2text.py@@ -15,6 +15,7 @@ jsonl = ("/nfs/beinian.lzr/workspace/GPT-4o/Data/Speech2Text/TestData/aishell1_test_speech2text.jsonl")output_dir = f"{os.path.join(ckpt_dir, ckpt_id)}"+device = "cuda:0"ckpt_dir = sys.argv[1]ckpt_id = sys.argv[2]diff --git a/examples/industrial_data_pretraining/llm_asr/demo_speech2text.sh b/examples/industrial_data_pretraining/llm_asr/demo_speech2text.shindex 4f521f286..81f211fa8 100644--- a/examples/industrial_data_pretraining/llm_asr/demo_speech2text.sh+++ b/examples/industrial_data_pretraining/llm_asr/demo_speech2text.sh@@ -1,10 +1,10 @@-ckpt_dir="/nfs/zhifu.gzf/ckpt/saves/qwen_1.5_7b/full/sft/asr_tts_text_exp1_ds_z3/checkpoint-11000"+ckpt_dir="/nfs/beinian.lzr/workspace/GPT-4o/Exp/exp6/5m-8gpu/exp6_speech2text_linear_ddp_0609"ckpt_id="model.pt.ep0.90000"jsonl_dir="/nfs/beinian.lzr/workspace/GPT-4o/Data/Speech2Text/TestData"-out_dir="${ckpt_dir}/asr"+out_dir="${ckpt_dir}/inference-${ckpt_id}"mkdir -p ${out_dir}device="cuda:0"@@ -12,9 +12,9 @@ device="cuda:0"for data_set in "librispeech_test_clean_speech2text.jsonl" "librispeech_test_other_speech2text.jsonl"; dojsonl=${jsonl_dir}/${data_set}output_dir=${out_dir}/${data_set}-- pred_file=${out_dir}/${data_set}/1best_recog/text_tn- ref_file=${out_dir}/${data_set}/1best_recog/label+ mkdir -p ${output_dir}+ pred_file=${output_dir}/${data_set}/1best_recog/text_tn+ ref_file=${output_dir}/${data_set}/1best_recog/labelpython ./demo_speech2text.py ${ckpt_dir} ${ckpt_id} ${jsonl} ${output_dir} ${device}@@ -26,9 +26,9 @@ donefor data_set in "aishell1_test_speech2text.jsonl" "aishell2_ios_test_speech2text.jsonl" "librispeech_test_other_speech2text.jsonl"; dojsonl=${jsonl_dir}/${data_set}output_dir=${out_dir}/${data_set}-- pred_file=${out_dir}/${data_set}/1best_recog/text_tn- ref_file=${out_dir}/${data_set}/1best_recog/label+ mkdir -p ${output_dir}+ pred_file=${output_dir}/${data_set}/1best_recog/text_tn+ ref_file=${output_dir}/${data_set}/1best_recog/labelpython ./demo_speech2text.py ${ckpt_dir} ${ckpt_id} ${jsonl} ${output_dir}@@ -39,9 +39,9 @@ donefor data_set in "s2tt_en2zh.v20240605.test.jsonl"; dojsonl=${jsonl_dir}/${data_set}output_dir=${out_dir}/${data_set}-- pred_file=${out_dir}/${data_set}/1best_recog/text_tn- ref_file=${out_dir}/${data_set}/1best_recog/label+ mkdir -p ${output_dir}+ pred_file=${output_dir}/${data_set}/1best_recog/text_tn+ ref_file=${output_dir}/${data_set}/1best_recog/labelpython ./demo_speech2text.py ${ckpt_dir} ${ckpt_id} ${jsonl} ${output_dir}@@ -52,9 +52,9 @@ donefor data_set in "s2tt_zh2en.v20240605.test.jsonl"; dojsonl=${jsonl_dir}/${data_set}output_dir=${out_dir}/${data_set}-- pred_file=${out_dir}/${data_set}/1best_recog/text_tn- ref_file=${out_dir}/${data_set}/1best_recog/label+ mkdir -p ${output_dir}+ pred_file=${output_dir}/${data_set}/1best_recog/text_tn+ ref_file=${output_dir}/${data_set}/1best_recog/labelpython ./demo_speech2text.py ${ckpt_dir} ${ckpt_id} ${jsonl} ${output_dir}diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.pyindex aacbe45a4..21072b0f7 100644--- a/funasr/models/llm_asr/model.py+++ b/funasr/models/llm_asr/model.py@@ -556,7 +556,7 @@ class LLMASR2(nn.Module):return contents- def data_load_speech(self, contents: dict, tokenizer, frontend, **kwargs):+ def data_load_speech(self, contents: dict, tokenizer, frontend, meta_data={}, **kwargs):system = contents["system"]user = contents["user"]@@ -594,7 +594,10 @@ class LLMASR2(nn.Module):)if sub_str.startswith("!"):try:+ time1 = time.perf_counter()data_src = load_audio_text_image_video(sub_str[1:], fs=frontend.fs)+ time2 = time.perf_counter()+ meta_data["load_data"] = f"{time2 - time1:0.3f}"except Exception as e:logging.error(f"Loading wav failed! {str(e)}, {traceback.format_exc()}")@@ -605,6 +608,15 @@ class LLMASR2(nn.Module):is_final=True,) # speech: [b, T, d]+ time3 = time.perf_counter()+ meta_data["extract_feat"] = f"{time3 - time2:0.3f}"+ meta_data["batch_data_time"] = (+ speech_lengths.sum().item()+ * frontend.frame_shift+ * frontend.lfr_n+ / 1000+ )+if kwargs.get("permute", True):speech = speech.permute(0, 2, 1)@@ -666,7 +678,7 @@ class LLMASR2(nn.Module):raise NotImplementedError("batch decoding is not implemented")contents = self.data_template(data_in[0])- output = self.data_load_speech(contents, tokenizer, frontend, **kwargs)+ output = self.data_load_speech(contents, tokenizer, frontend, meta_data=meta_data, **kwargs)batch = to_device(output, kwargs["device"])# audio encoder