File size: 3,407 Bytes
b815c4a
a5753ad
4841807
f427fe9
b815c4a
f427fe9
 
b815c4a
 
 
 
 
 
 
 
f427fe9
b815c4a
 
 
 
 
 
 
 
f427fe9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a5753ad
 
 
 
 
 
 
 
b815c4a
f427fe9
 
 
 
b815c4a
f427fe9
 
 
 
 
 
 
 
b815c4a
f427fe9
b815c4a
 
 
4841807
 
 
 
 
 
 
 
a5753ad
b815c4a
 
 
 
 
 
 
f427fe9
 
 
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
import torch
import pickle
import whisper
import streamlit as st
import torchaudio as ta

from io import BytesIO
from transformers import AutoProcessor, SeamlessM4TModel, WhisperProcessor, WhisperForConditionalGeneration

if torch.cuda.is_available():
    device = "cuda:0"
    torch_dtype = torch.float16
else:
    device = "cpu"
    torch_dtype = torch.float32

SAMPLING_RATE=16000
task = "transcribe"

print(f"{device} Active!")

# load Whisper model and processor
processor = WhisperProcessor.from_pretrained("openai/whisper-small")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")

# Title of the app
st.title("Audio Player with Live Transcription")

# Sidebar for file uploader and submit button
st.sidebar.header("Upload Audio Files")
uploaded_files = st.sidebar.file_uploader("Choose audio files", type=["mp3", "wav"], accept_multiple_files=True)
submit_button = st.sidebar.button("Submit")


# def transcribe_audio(audio_data):
#     recognizer = sr.Recognizer()
#     with sr.AudioFile(audio_data) as source:
#         audio = recognizer.record(source)
#     try:
#         # Transcribe the audio using Google Web Speech API
#         transcription = recognizer.recognize_google(audio)
#         return transcription
#     except sr.UnknownValueError:
#         return "Unable to transcribe the audio."
#     except sr.RequestError as e:
#         return f"Could not request results; {e}"

def detect_language(audio_file):
    whisper_model = whisper.load_model("base")
    mel = whisper.log_mel_spectrogram(trimmed_audio).to(whisper_model.device)
    # detect the spoken language
    _, probs = whisper_model.detect_language(mel)
    print(f"Detected language: {max(probs[0], key=probs[0].get)}")
    return max(probs[0], key=probs[0].get)
    
if submit_button and uploaded_files is not None:
    st.write("Files uploaded successfully!")

    for uploaded_file in uploaded_files:
        # Display file name and audio player

        st.write(f"**File name**: {uploaded_file.name}")
        st.audio(uploaded_file, format=uploaded_file.type)

        # Transcription section
        st.write("**Transcription**:")

        # Read the uploaded file data
        waveform, sampling_rate = ta.load(uploaded_file.getvalue())
        resampled_inp = ta.functional.resample(waveform, orig_freq=sampling_rate, new_freq=SAMPLING_RATE)

        input_features = processor(resampled_inp[0], sampling_rate=16000, return_tensors='pt').input_features

        if task == "translate":
            
            # Detect Language 
            lang = detect_language(input_features)
            with open('languages.pkl', 'rb') as f:
                lang_dict = pickle.load(f)
            detected_language = lang_dict[lang]

            # Set decoder & Predict translation
            forced_decoder_ids = processor.get_decoder_prompt_ids(language=detected_language, task="translate")
            predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids)
        else:
            predicted_ids = model.generate(input_features)
        # decode token ids to text
        transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
        st.write(transcription)
        # print(waveform, sampling_rate)
        # Run transcription function and display
        # import pdb;pdb.set_trace()
        # st.write(audio_data.getvalue())