File size: 4,240 Bytes
32b6530 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 |
import torch
import gradio as gr
import time
import numpy as np
import scipy.io.wavfile
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
# ✅ 1️⃣ Force Model to Run on CPU
device = "cpu"
torch_dtype = torch.float32 # Use CPU-friendly float type
MODEL_NAME = "openai/whisper-tiny" # ✅ Switched to smallest model for fastest performance
# ✅ 2️⃣ Load Whisper Tiny Model on CPU
model = AutoModelForSpeechSeq2Seq.from_pretrained(
MODEL_NAME, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
)
model.to(device)
# ✅ 3️⃣ Load Processor & Pipeline
processor = AutoProcessor.from_pretrained(MODEL_NAME)
pipe = pipeline(
task="automatic-speech-recognition",
model=model,
tokenizer=processor.tokenizer,
feature_extractor=processor.feature_extractor,
chunk_length_s=2, # ✅ Process in 2-second chunks for ultra-low latency
torch_dtype=torch_dtype,
device=device,
)
# ✅ 4️⃣ Real-Time Streaming Transcription (Microphone)
def stream_transcribe(stream, new_chunk):
start_time = time.time()
try:
sr, y = new_chunk
# ✅ Convert stereo to mono
if y.ndim > 1:
y = y.mean(axis=1)
y = y.astype(np.float32)
y /= np.max(np.abs(y))
# ✅ Append to Stream
if stream is not None:
stream = np.concatenate([stream, y])
else:
stream = y
# ✅ Run Transcription
transcription = pipe({"sampling_rate": sr, "raw": stream})["text"]
latency = time.time() - start_time
return stream, transcription, f"{latency:.2f} sec"
except Exception as e:
print(f"Error: {e}")
return stream, str(e), "Error"
# ✅ 5️⃣ Transcription for File Upload
def transcribe(inputs, previous_transcription):
start_time = time.time()
try:
# ✅ Convert file input to correct format
sample_rate, audio_data = inputs
transcription = pipe({"sampling_rate": sample_rate, "raw": audio_data})["text"]
previous_transcription += transcription
latency = time.time() - start_time
return previous_transcription, f"{latency:.2f} sec"
except Exception as e:
print(f"Error: {e}")
return previous_transcription, "Error"
# ✅ 6️⃣ Clear Function
def clear():
return ""
# ✅ 7️⃣ Gradio Interface (Microphone Streaming)
with gr.Blocks() as microphone:
gr.Markdown(f"# Whisper Tiny - Real-Time Transcription (CPU) 🎙️")
gr.Markdown(f"Using [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) for ultra-fast speech-to-text.")
with gr.Row():
input_audio_microphone = gr.Audio(sources=["microphone"], type="numpy", streaming=True)
output = gr.Textbox(label="Live Transcription", value="")
latency_textbox = gr.Textbox(label="Latency (seconds)", value="0.0")
with gr.Row():
clear_button = gr.Button("Clear Output")
state = gr.State()
input_audio_microphone.stream(
stream_transcribe, [state, input_audio_microphone],
[state, output, latency_textbox], time_limit=30, stream_every=1
)
clear_button.click(clear, outputs=[output])
# ✅ 8️⃣ Gradio Interface (File Upload)
with gr.Blocks() as file:
gr.Markdown(f"# Upload Audio File for Transcription 🎵")
gr.Markdown(f"Using [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) for speech-to-text.")
with gr.Row():
input_audio = gr.Audio(sources=["upload"], type="numpy")
output = gr.Textbox(label="Transcription", value="")
latency_textbox = gr.Textbox(label="Latency (seconds)", value="0.0")
with gr.Row():
submit_button = gr.Button("Submit")
clear_button = gr.Button("Clear Output")
submit_button.click(transcribe, [input_audio, output], [output, latency_textbox])
clear_button.click(clear, outputs=[output])
# ✅ 9️⃣ Final Gradio App (Supports Microphone & File Upload)
with gr.Blocks(theme=gr.themes.Ocean()) as demo:
gr.TabbedInterface([microphone, file], ["Microphone", "Upload Audio"])
# ✅ 1️⃣0️⃣ Run Gradio Locally
if __name__ == "__main__":
demo.launch()
|