app.py

import gradio as gr
from transformers import pipeline
from gradio_client import Client, file
import json

language_classifier = Client("adrien-alloreview/speechbrain-lang-id-voxlingua107-ecapa")
transcriber = Client("tensorlake/audio-extractors")
emotion_detector = pipeline(
    "audio-classification",
    model="HowMannyMore/wav2vec2-lg-xlsr-ur-speech-emotion-recognition",
)
model_name_rus = "IlyaGusev/rubertconv_toxic_clf"
toxic_detector = pipeline(
    "text-classification",
    model=model_name_rus,
    tokenizer=model_name_rus,
    framework="pt",
    max_length=512,
    truncation=True,
)


def detect_language(file_path):
    try:
        result = language_classifier.predict(param_0=file(file_path), api_name="/predict")
        language_result = result["label"].split(": ")[1]
        if language_result.lower() in ["russian", "belarussian", "ukrainian"]:
            selected_language = "russian"
        else:
            selected_language = "kazakh"
        return selected_language
    except Exception as e:
        print(f"Language detection failed: {e}")
        return None


def request_gradio(file_path, language):
    try:
        result = transcriber.predict(
            audio_filepath=file(file_path),
            task="transcribe",
            batch_size=24,
            chunk_length_s=30,
            sampling_rate=16000,
            language=language,
            num_speakers=2,
            min_speakers=2,
            max_speakers=2,
            assisted=False,
            api_name="/transcribe",
        )
        return result
    except Exception as e:
        print(f"Transcription failed: {e}")
        return f"Transcription failed: {e}"


def detect_emotion(audio):
    try:
        res = emotion_detector(audio)
        emotion_with_max_score = res[0]["label"]
        return emotion_with_max_score
    except Exception as e:
        return f"Emotion detection failed: {e}"


def detect_toxic_local(text_whisper):
    try:
        res = toxic_detector([text_whisper])[0]["label"]
        if res == "toxic":
            return True
        elif res == "neutral":
            return False
        else:
            return None
    except Exception as e:
        print(f"Toxicity detection failed: {e}")
        return None


def assessment(file_path, language):
    language = detect_language(file_path) or "unknown"
    result_text = request_gradio(file_path, language) or ""
    result_emotion = detect_emotion(file_path) or "unknown"
    result_toxic = detect_toxic_local(result_text) or False
    return json.dumps({"transcription": result_text, "emotion": result_emotion, "toxic": result_toxic,})

demo = gr.Blocks()

with demo:
    with gr.Tabs():
        with gr.TabItem('Language Detection'):
            language_detection_interface = gr.Interface(
                fn=detect_language,
                inputs=gr.Audio(sources=["upload"], type="filepath"),
                outputs='text',
            )
        with gr.TabItem('Toxic & Emotion Detection'):
            toxic_and_emotion_detection_interface = gr.Interface(
                fn=assessment,
                inputs=[gr.Audio(sources=["upload"], type="filepath"), gr.Textbox(label="Language")],
                outputs='json',
            )
demo.launch()