app.py

import torch
#from transformers.pipelines.audio_utils import ffmpeg_read
from speechscore import SpeechScore 
import gradio as gr
import pprint

MODEL_NAME = "alibabasglab/speechscore"
BATCH_SIZE = 1

device = 0 if torch.cuda.is_available() else "cpu"

def score_file(test_file, ref_file, score_list_nis, score_list_is):
    score_list = []
    if score_list_nis is not None:
        score_list += score_list_nis
    if score_list_is is not None:
        score_list += score_list_is        
    mySpeechScore = SpeechScore(score_list)  
    scores = mySpeechScore(test_path=test_file, reference_path=ref_file, window=None, score_rate=16000, return_mean=False)
    return scores

def score_mic(test_file, score_list_nis):
    score_list = []
    if score_list_nis is not None:
        score_list += score_list_nis       
    mySpeechScore = SpeechScore(score_list)  
    scores = mySpeechScore(test_path=test_file, reference_path=None, window=None, score_rate=16000, return_mean=False)
    return scores

demo = gr.Blocks()

file_score = gr.Interface(
    fn=score_file,
    inputs=[
        gr.Audio(sources=["upload"], label="test file", type="filepath"),
        gr.Audio(sources=["upload"], label="reference file", type="filepath"),
        gr.Dropdown(
            ["DNSMOS", "SRMR"], value=["DNSMOS"], multiselect=True, label="Non-Intrusive Scores", info="Choose scores to include, no reference audio is required."
        ),
        gr.Dropdown(
            ["PESQ", 'NB_PESQ', 'STOI', 'SISDR',
        'FWSEGSNR', 'LSD', 'BSSEval', 'DNSMOS',
        'SNR', 'SSNR', 'LLR', 'CSIG', 'CBAK',
        'COVL', 'MCD'], value=["PESQ", "STOI"], multiselect=True, label="Intrusive Scores", info="Choose scores to include, reference audio is required."
        ),
    ],
    outputs="text",
    title="Score speech quality for an audio clip",
    description=(
        "Score speech quality with the click of a button! Demo includes the"
        " commonly used speech quality assessments for the audio file"
        " of arbitrary length."
    ),
)

mic_score = gr.Interface(
    fn=score_mic,
    inputs=[
        gr.Audio(sources=["microphone"],
                waveform_options=gr.WaveformOptions(
                waveform_color="#01C6FF",
                waveform_progress_color="#0066B4",
                skip_length=2,
                show_controls=False,                
                ),
                type='numpy',
            ),
        gr.Dropdown(
            ["DNSMOS", "SRMR"], value=["DNSMOS", "SRMR"], multiselect=True, label="Non-Intrusive Scores", info="Choose scores to include, no reference audio is required."
        ),
    ],
    outputs="text",
    title="Test microphone quality using speech score",
    description=(
        "Score your microphone quality with the click of a button!"
        " Uses the most popular method to test your microphone quality"
        " with a short speech clip."
    ),
)

with demo:
    gr.TabbedInterface([mic_score, file_score], ["Score Microphone Quality", "Score Speech Quality"])

demo.launch()