File size: 2,365 Bytes
a934af5
 
 
 
 
b4119c8
a934af5
b4119c8
a934af5
 
b3a902e
a934af5
 
 
 
 
 
b4119c8
a934af5
b3a902e
a934af5
b4119c8
 
 
a934af5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import gradio as gr
import torch
import torchaudio
import numpy as np
import threading
import queue
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import sounddevice as sd
import tempfile
import wave

# Load ASR Model
model_name = "Futuresony/Future-sw_ASR-24-02-2025"
processor = Wav2Vec2Processor.from_pretrained(model_name)
model = Wav2Vec2ForCTC.from_pretrained(model_name)

# Streaming Variables
q = queue.Queue()
streaming = True

# Function to Record Audio in Chunks
def callback(indata, frames, time, status):
    if status:
        print(status)
    q.put(indata.copy())

# Function to Continuously Transcribe Audio
def transcribe_stream():
    global streaming
    samplerate = 16000  # Model expects 16kHz audio

    # Start recording stream
    with sd.InputStream(samplerate=samplerate, channels=1, callback=callback):
        while streaming:
            audio_data = []
            
            try:
                # Collect small audio chunks from the queue
                for _ in range(5):  # Adjust to control update frequency
                    audio_chunk = q.get(timeout=1)
                    audio_data.append(audio_chunk)
                    
                # Convert recorded chunks to numpy array
                audio_np = np.concatenate(audio_data, axis=0).flatten()
                
                # Process & transcribe
                input_values = processor(audio_np, sampling_rate=16000, return_tensors="pt").input_values
                with torch.no_grad():
                    logits = model(input_values).logits
                predicted_ids = torch.argmax(logits, dim=-1)
                transcription = processor.batch_decode(predicted_ids)[0]

                yield transcription  # Stream output live

            except queue.Empty:
                continue

# Gradio Live Interface
def live_transcription():
    return transcribe_stream()

interface = gr.Interface(
    fn=live_transcription,
    inputs=None,
    outputs=gr.Textbox(label="Live Transcription"),
    live=True,
    title="Swahili Live Streaming ASR",
    description="Speak continuously, and the subtitles will appear in real-time.",
)

# Run Transcription in Background Thread
thread = threading.Thread(target=transcribe_stream)
thread.daemon = True
thread.start()

# Launch Gradio App
if __name__ == "__main__":
    interface.launch()