import gradio as gr
import librosa
from asr import transcribe, ASR_EXAMPLES, ASR_LANGUAGES, ASR_NOTE
from tts import synthesize, TTS_EXAMPLES, TTS_LANGUAGES
from lid import identify, LID_EXAMPLES
demo = gr.Blocks()
mms_source_trans = gr.Audio()
mms_transcribe = gr.Interface(
fn=transcribe,
inputs=[
mms_source_trans,
gr.Dropdown(
[f"{k} ({v})" for k, v in ASR_LANGUAGES.items()],
label="Language",
value="eng English",
),
# gr.Checkbox(label="Use Language Model (if available)", default=True),
],
outputs="text",
examples=ASR_EXAMPLES,
title="Speech-to-text",
description=(
"Transcribe audio from a microphone or input file in your desired language."
),
article=ASR_NOTE,
allow_flagging="never",
)
mms_synthesize = gr.Interface(
fn=synthesize,
inputs=[
gr.Text(label="Input text"),
gr.Dropdown(
[f"{k} ({v})" for k, v in TTS_LANGUAGES.items()],
label="Language",
value="eng English",
),
gr.Slider(minimum=0.1, maximum=4.0, value=1.0, step=0.1, label="Speed"),
],
outputs=[
gr.Audio(label="Generated Audio", type="numpy"),
gr.Text(label="Filtered text after removing OOVs"),
],
examples=TTS_EXAMPLES,
title="Text-to-speech",
description=("Generate audio in your desired language from input text."),
allow_flagging="never",
)
mms_source_iden = gr.Audio()
mms_identify = gr.Interface(
fn=identify,
inputs=[
mms_source_iden,
],
outputs=gr.Label(num_top_classes=10),
examples=LID_EXAMPLES,
title="Language Identification",
description=("Identity the language of input audio."),
allow_flagging="never",
)
# tabbed_interface = gr.TabbedInterface(
# [mms_transcribe, mms_synthesize, mms_identify],
# ["Speech-to-text", "Text-to-speech", "Language Identification"],
# )
tabbed_interface = gr.TabbedInterface(
[ mms_identify],
["Language Identification"],
)
with gr.Blocks() as demo:
gr.Markdown(
"MMS: Scaling Speech Technology to 1000+ languages demo. See our blog post and paper.
"
)
gr.HTML(
"""Click on the appropriate tab to explore Speech-to-text (ASR), Text-to-speech (TTS) and Language identification (LID) demos. """
)
gr.HTML(
"""You can also finetune MMS models on your data using the recipes provides here - ASR TTS """
)
gr.HTML(
"""
for more control and no queue."""
)
tabbed_interface.render()
gr.HTML(
"""
"""
)
demo.queue(concurrency_count=3)
demo.launch()