|
import nemo |
|
from nemo.collections.asr.models.msdd_models import NeuralDiarizer |
|
import gradio as gr |
|
import pandas as pd |
|
import torch |
|
|
|
device = "cuda" if torch.cuda.is_available() else "cpu" |
|
|
|
model = NeuralDiarizer.from_pretrained("diar_msdd_telephonic").to(device) |
|
|
|
def run_diarization(path1): |
|
annotation = model(path1) |
|
rttm=annotation.to_rttm() |
|
df = pd.DataFrame(columns=['start_time', 'end_time', 'speaker']) |
|
for idx,line in enumerate(rttm.splitlines()): |
|
split = line.split() |
|
start_time, duration, speaker = split[3], split[4], split[7] |
|
end_time = float(start_time) + float(duration) |
|
df.loc[idx] = start_time, end_time, speaker |
|
return df |
|
|
|
inputs = [ |
|
gr.components.Audio(source="microphone", type="filepath", optional=True, label="Input Audio"), |
|
] |
|
output = gr.components.Dataframe() |
|
|
|
description = ( |
|
"This demonstration will perform offline speaker diarization on an audio file using nemo" |
|
) |
|
|
|
article = ( |
|
"<p style='text-align: center'>" |
|
"<a href='https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/diar_msdd_telephonic' target='_blank'>ποΈ Learn more about MSDD model</a> | " |
|
"<a href='https://arxiv.org/abs/2203.15974' target='_blank'>π MSDD paper</a> | " |
|
"<a href='https://github.com/NVIDIA/NeMo' target='_blank'>π§βπ» Repository</a>" |
|
"</p>" |
|
) |
|
examples = [ |
|
["data/sample_interview_conversation.wav"], |
|
["data/id10270_5r0dWxy17C8-00001.wav"], |
|
] |
|
|
|
interface = gr.Interface( |
|
fn=run_diarization, |
|
inputs=inputs, |
|
outputs=output, |
|
title="Offline Speaker Diarization with NeMo", |
|
description=description, |
|
article=article, |
|
layout="horizontal", |
|
theme="huggingface", |
|
allow_flagging=False, |
|
live=False, |
|
examples=examples, |
|
) |
|
interface.launch(enable_queue=True) |
|
|