app.py

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

import datetime
import math
import io
import os
import tempfile
import json
from typing import Optional

from pyannote.audio import Audio, Pipeline
from pyannote.core import Segment

import torch
import gradio as gr

from config import model_config




device = "cuda:0" if torch.cuda.is_available() else "cpu"
model_dir = snapshot_download(model_config['model_dir'])

model = AutoModel(
    model=model_dir,
    trust_remote_code=False,
    remote_code="./model.py",
    vad_model="fsmn-vad",
    punc_model="ct-punc", 
    vad_kwargs={"max_single_segment_time": 30000},
    ncpu=torch.get_num_threads(),
    batch_size=1,
    hub="hf",
    device=device,
)

def transcribe_audio(file_path, vad_model="fsmn-vad", punc_model="ct-punc", vad_kwargs='{"max_single_segment_time": 30000}', 
                     batch_size=1, language="auto", use_itn=True, batch_size_s=60, 
                     merge_vad=True, merge_length_s=15, batch_size_threshold_s=50, 
                     hotword=" ", spk_model="cam++", ban_emo_unk=True):
    try:
        vad_kwargs = json.loads(vad_kwargs)
        
        temp_file_path = file_path

        res = model.generate(
            input=temp_file_path,
            cache={},
            language=language,
            use_itn=use_itn,
            batch_size_s=batch_size_s,
            merge_vad=merge_vad,
            merge_length_s=merge_length_s,
            batch_size_threshold_s=batch_size_threshold_s,
            hotword=hotword,
            spk_model=spk_model,
            ban_emo_unk=ban_emo_unk
        )

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

    except Exception as e:
        return str(e)

inputs = [
    gr.Audio(type="filepath"),
    gr.Textbox(value="fsmn-vad", label="VAD Model"),
    gr.Textbox(value="ct-punc", label="PUNC Model"),
    gr.Textbox(value='{"max_single_segment_time": 30000}', label="VAD Kwargs"),
    gr.Slider(1, 10, value=1, step=1, label="Batch Size"),
    gr.Textbox(value="auto", label="Language"),
    gr.Checkbox(value=True, label="Use ITN"),
    gr.Slider(30, 120, value=60, step=1, label="Batch Size (seconds)"),
    gr.Checkbox(value=True, label="Merge VAD"),
    gr.Slider(5, 60, value=15, step=1, label="Merge Length (seconds)"),
    gr.Slider(10, 100, value=50, step=1, label="Batch Size Threshold (seconds)"),
    gr.Textbox(value=" ", label="Hotword"),
    gr.Textbox(value="cam++", label="Speaker Model"),
    gr.Checkbox(value=True, label="Ban Emotional Unknown"),
]

outputs = gr.Textbox(label="Transcription")

gr.Interface(
    fn=transcribe_audio, 
    inputs=inputs, 
    outputs=outputs, 
    title="ASR Transcription with FunASR"
).launch()