sensevoice

游雁

2024-07-16 8b74979bd38c70c731e04b32308f2c9edeabdc37

sensevoice

 examples/industrial_data_pretraining/sense_voice/demo.py

@@ -3,26 +3,82 @@
# Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
#  MIT License  (https://opensource.org/licenses/MIT)

import sys

from funasr import AutoModel
from funasr.utils.postprocess_utils import rich_transcription_postprocess

model_dir = "iic/SenseVoiceSmall"
input_file = (
    "https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav"
)


model = AutoModel(
    model=model_dir,
    vad_model="fsmn-vad",
    vad_kwargs={"max_single_segment_time": 30000},
    device="cpu",
)

# en
res = model.generate(
    input=input_file,
    input=f"{model.model_path}/example/en.mp3",
    cache={},
    language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"
    use_itn=False,
    use_itn=True,
    batch_size_s=60,
    merge_vad=True,  #
    merge_length_s=15,
)

text = rich_transcription_postprocess(res[0]["text"])
print(text)

# zh
res = model.generate(
    input=f"{model.model_path}/example/zh.mp3",
    cache={},
    language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"
    use_itn=True,
    batch_size_s=60,
    merge_vad=True,  #
    merge_length_s=15,
)
text = rich_transcription_postprocess(res[0]["text"])
print(text)

# yue
res = model.generate(
    input=f"{model.model_path}/example/yue.mp3",
    cache={},
    language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"
    use_itn=True,
    batch_size_s=60,
    merge_vad=True,  #
    merge_length_s=15,
)
text = rich_transcription_postprocess(res[0]["text"])
print(text)

# ja
res = model.generate(
    input=f"{model.model_path}/example/ja.mp3",
    cache={},
    language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"
    use_itn=True,
    batch_size_s=60,
    merge_vad=True,  #
    merge_length_s=15,
)
text = rich_transcription_postprocess(res[0]["text"])
print(text)


# ko
res = model.generate(
    input=f"{model.model_path}/example/ko.mp3",
    cache={},
    language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"
    use_itn=True,
    batch_size_s=60,
    merge_vad=True,  #
    merge_length_s=15,
)
text = rich_transcription_postprocess(res[0]["text"])
print(text)

 funasr/auto/auto_model.py

@@ -339,7 +339,9 @@
        #  FIX(gcf): concat the vad clips for sense vocie model for better aed
        if kwargs.get("merge_vad", False):
            for i in range(len(res)):
                res[i]["value"] = merge_vad(res[i]["value"], kwargs.get("merge_length", 15000))
                res[i]["value"] = merge_vad(
                    res[i]["value"], kwargs.get("merge_length_s", 15) * 1000
                )

        # step.2 compute asr model
        model = self.model
@@ -380,6 +382,9 @@
            if len(sorted_data) > 0 and len(sorted_data[0]) > 0:
                batch_size = max(batch_size, sorted_data[0][0][1] - sorted_data[0][0][0])

            if kwargs["device"] == "cpu":
                batch_size = 0

            beg_idx = 0
            beg_asr_total = time.time()
            time_speech_total_per_sample = speech_lengths / 16000