Add-Vishnu's picture
Commented base model and using medium model
cd321a3
raw
history blame
2.04 kB
import gradio as gr
import soundfile as sf
import tempfile
import shutil
import os
import librosa
import time
import numpy as np
import subprocess
from pywhispercpp.model import Model
# model = Model('base.en', n_threads=6,models_dir="./Models") # Only english
# model = Model('base', n_threads=6,models_dir="./Models",language="hindi",translate=False) # Multilingual
model = Model('medium-q5_0', n_threads=6,models_dir="./Models",language="hindi",translate=False) # Multilingual
def resample_to_16k(audio, orig_sr):
y_resampled = librosa.resample(y=audio, orig_sr=orig_sr, target_sr = 16000)
return y_resampled
def transcribe(audio):
print(type(audio))
sr,y = audio
y = y.astype(np.float32)
y /= np.max(np.abs(y))
y_resampled = resample_to_16k(y, sr)
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_audio:
temp_audio_path = temp_audio.name
sf.write(temp_audio_path, y_resampled, 16000)
start_time_py = time.time()
py_result = model.transcribe(f'{temp_audio_path}', n_threads=6)
end_time_py = time.time()
print("Py_result : ",py_result)
print("--------------------------")
print(f"Execution time using py: {end_time_py - start_time_py} seconds")
output_text = ""
for segment in py_result:
output_text+=segment.text
return output_text, (end_time_py - start_time_py)
demo = gr.Interface(
transcribe,
inputs = "microphone",
# gr.Audio(sources=["microphone"]),
outputs=[gr.Textbox(label="Py_Transcription"),gr.Textbox(label="Time taken for Transcription")],
examples=["./Samples/Hindi_1.mp3","./Samples/Hindi_2.mp3","./Samples/Tamil_1.mp3","./Samples/Tamil_2.mp3","./Samples/Marathi_1.mp3","./Samples/Marathi_2.mp3","./Samples/Nepal_1.mp3","./Samples/Nepal_2.mp3","./Samples/Telugu_1.wav","./Samples/Telugu_2.wav","./Samples/Malayalam_1.wav","./Samples/Malayalam_2.wav","./Samples/Gujarati_1.wav","./Samples/Gujarati_2.wav","./Samples/Bengali_1.wav","./Samples/Bengali_2.wav"]
)
demo.launch()