# Real-time speech processing

A multilingual text-to-text model preloaded with meta-llama/Llama-3.2-1B-Instruct weights

Ultravox is a multilingual audio-to-text model optimized based on the LLaMA-3-2.1B architecture, supporting speech recognition and transcription tasks in multiple languages.

This is an audio segmentation model capable of detecting speaker changes, voice activity, and overlapping speech, suitable for audio analysis in multi-speaker scenarios.

A fine-tuned Uzbek speech recognition model based on Oyqiz/uzbek_stt, specifically optimized for legal and military domain data

This is a speaker segmentation model based on pyannote.audio, capable of detecting speech activity, speaker changes, and overlapping speech.

Speaker diarization pipeline trained on pyannote.audio 3.0.0, supporting automatic voice activity detection, speaker change detection and overlapping speech detection

A speech recognition model fine-tuned on the speech_commands dataset based on facebook/wav2vec2-base, achieving 97.35% accuracy

A fine-tuned voice command classification model based on facebook/wav2vec2-base, trained on the superb dataset with an accuracy of 98.09%

An end-to-end speech translation model based on S2T architecture, specifically designed for English-to-Dutch speech translation tasks

This is a Greek automatic speech recognition model based on the XLSR-Wav2Vec2 architecture, developed by the Hellenic Military Academy and the Technical University of Crete.

A toolkit for speech enhancement (denoising) using the SepFormer model, pre-trained on the WHAM! dataset (8kHz sampling rate version) to remove environmental noise and reverberation.

This model achieves speech enhancement (denoising + dereverberation) through the SepFormer architecture, pre-trained on the WHAMR! dataset (8kHz), with a test set SI-SNR of 10.59dB.

A speech-to-text transformer model for end-to-end English to Spanish speech translation

A ConvTasNet model trained based on the Asteroid framework, designed to separate 3 independent audio sources from mixed audio, specifically optimized for noisy speech data at 8kHz sampling rate.