update

speech_asr

2023-03-15 b4598f30a54c3a8d5e6084d983fac0fa5a51992b

update

 egs_modelscope/speaker_diarization/speech_diarization_eend-ola-en-us-callhome-8k/infer.py

@@ -2,8 +2,9 @@
from modelscope.utils.constant import Tasks

inference_diar_pipline = pipeline(
    task=Tasks.speaker_diarization,
    task=Tasks.auto_speech_recognition,
    model='damo/speech_diarization_eend-ola-en-us-callhome-8k',
    model_revision="v1.0.0",
)
results = inference_diar_pipline(audio_in=["https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_data/record.wav"])
results = inference_diar_pipline(audio_in=["https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_data/record2.wav"])
print(results)

 funasr/bin/asr_inference_launch.py

@@ -234,6 +234,9 @@
    elif mode == "rnnt":
        from funasr.bin.asr_inference_rnnt import inference_modelscope
        return inference_modelscope(**kwargs)
    elif mode == "eend-ola":
        from funasr.bin.eend_ola_inference import inference_modelscope
        return inference_modelscope(mode=mode, **kwargs)
    else:
        logging.info("Unknown decoding mode: {}".format(mode))
        return None

 funasr/bin/eend_ola_inference.py

@@ -16,8 +16,8 @@

import numpy as np
import torch
from typeguard import check_argument_types
from scipy.signal import medfilt
from typeguard import check_argument_types

from funasr.models.frontend.wav_frontend import WavFrontendMel23
from funasr.tasks.diar import EENDOLADiarTask
@@ -27,6 +27,7 @@
from funasr.utils.types import str2bool
from funasr.utils.types import str2triple_str
from funasr.utils.types import str_or_none


class Speech2Diarization:
    """Speech2Diarlization class
@@ -237,7 +238,7 @@
            results = speech2diar(**batch)

            # post process
            a = results[0].cpu().numpy()
            a = results[0][0].cpu().numpy()
            a = medfilt(a, (11, 1))
            rst = []
            for spkid, frames in enumerate(a.T):
@@ -246,8 +247,8 @@
                fmt = "SPEAKER {:s} 1 {:7.2f} {:7.2f} <NA> <NA> {:s} <NA>"
                for s, e in zip(changes[::2], changes[1::2]):
                    st = s / 10.
                    ed = e / 10.
                    rst.append(fmt.format(keys[0], st, ed, "{}_{}".format(keys[0],str(spkid))))
                    dur = (e - s) / 10.
                    rst.append(fmt.format(keys[0], st, dur, "{}_{}".format(keys[0], str(spkid))))

            # Only supporting batch_size==1
            value = "\n".join(rst)