sanjitaa commited on
Commit
3023ee7
β€’
1 Parent(s): e3f88a8

upload app files

Browse files
Files changed (3) hide show
  1. app.py +90 -0
  2. packages.txt +1 -0
  3. requirements.txt +3 -0
app.py ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+
3
+ import gradio as gr
4
+ import yt_dlp as youtube_dl
5
+ from transformers import pipeline
6
+ from transformers.pipelines.audio_utils import ffmpeg_read
7
+ from transformers import MBartForConditionalGeneration, MBart50TokenizerFast
8
+
9
+
10
+ import tempfile
11
+ import os
12
+
13
+ MODEL_NAME = "openai/whisper-medium"
14
+ BATCH_SIZE = 8
15
+ FILE_LIMIT_MB = 1000
16
+
17
+ device = 0 if torch.cuda.is_available() else "cpu"
18
+
19
+ pipe = pipeline(
20
+ task="automatic-speech-recognition",
21
+ model=MODEL_NAME,
22
+ chunk_length_s=30,
23
+ device=device,
24
+ )
25
+
26
+ model = MBartForConditionalGeneration.from_pretrained("sanjitaa/mbart-many-to-many")
27
+ tokenizer = MBart50TokenizerFast.from_pretrained("sanjitaa/mbart-many-to-many")
28
+
29
+ def translate(inputs, task):
30
+ if inputs is None:
31
+ raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
32
+
33
+ text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"]
34
+
35
+ encoded_text = tokenizer(text, return_tensors="pt")
36
+ tokenizer.src_lang = "en_XX"
37
+
38
+ generated_tokens = model.generate(
39
+ **encoded_text,
40
+ forced_bos_token_id=tokenizer.lang_code_to_id["fr_XX"]
41
+ )
42
+ result = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
43
+ return result
44
+
45
+
46
+ demo = gr.Blocks()
47
+
48
+ mf_transcribe = gr.Interface(
49
+ fn=translate,
50
+ inputs=[
51
+ gr.inputs.Audio(source="microphone", type="filepath", optional=True),
52
+ gr.inputs.Radio(["translate"], label="Task", default="translate"),
53
+
54
+ ],
55
+ outputs="text",
56
+ layout="horizontal",
57
+ theme="huggingface",
58
+ title="Whisper Medium: Transcribe Audio",
59
+ description=(
60
+ "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
61
+ f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and πŸ€— Transformers to transcribe audio files"
62
+ " of arbitrary length."
63
+ ),
64
+ allow_flagging="never",
65
+ )
66
+
67
+ file_transcribe = gr.Interface(
68
+ fn=transcribe,
69
+ inputs=[
70
+ gr.inputs.Audio(source="upload", type="filepath", optional=True, label="Audio file"),
71
+ gr.inputs.Radio(["translate"], label="Task", default="transcribe"),
72
+ ],
73
+ outputs="text",
74
+ layout="horizontal",
75
+ theme="huggingface",
76
+ title="Whisper Large V2: Transcribe Audio",
77
+ description=(
78
+ "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
79
+ f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and πŸ€— Transformers to transcribe audio files"
80
+ " of arbitrary length."
81
+ ),
82
+ allow_flagging="never",
83
+ )
84
+
85
+
86
+ with demo:
87
+ gr.TabbedInterface([mf_transcribe, file_transcribe], ["Microphone", "Audio file"])
88
+
89
+ demo.launch(enable_queue=True)
90
+
packages.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ ffmpeg
requirements.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ git+https://github.com/huggingface/transformers
2
+ torch
3
+ yt-dlp