app.py

import gradio as gr
# from transformers import WhisperProcessor, WhisperForConditionalGeneration
from datasets import load_dataset
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline

import torch
# import librosa

# 加载 Whisper 模型和 processor
# model_name = "openai/whisper-large-v3-turbo"
# processor = WhisperProcessor.from_pretrained(model_name)
# model = WhisperForConditionalGeneration.from_pretrained(model_name)

model_id = "openai/whisper-large-v3-turbo"

models = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id,  low_cpu_mem_usage=True
)

model = pipeline("automatic-speech-recognition", model=models, chunk_length_s=30, device=0)

# 加载数据集 bigcode/the-stack

ds = load_dataset("CoIR-Retrieval/CodeSearchNet-php-queries-corpus")

def transcribe(audio_path):
    # 加载音频文件并转换为信号
    # audio, sr = librosa.load(audio_path, sr=16000)
    # input_values = processor(audio_path, return_tensors="pt", sampling_rate=16000).["text"]

    # # 模型推理
    # with torch.no_grad():
    #     logits = model(input_values).logits

    # predicted_ids = torch.argmax(logits, dim=-1)
    # transcription = processor.batch_decode(predicted_ids)

    transcription = model(audio_path,batch_size=1000, generate_kwargs={"task": "transcribe"}, return_timestamps=True)["text"]

    # result = pipe(sample)
    # 返回转录结果
    return transcription

    
# Gradio 界面
iface = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(sources="microphone",  type="filepath"),
    outputs="text",
    title="Whisper Transcription for Developers",
    description="使用 Whisper 和 bigcode 数据集转录开发者相关术语。"
)

# 启动 Gradio 应用
iface.launch()