Popular GitHub Repositories for Wer

Attempt at tracking states of the arts and recent results (bibliography) on speech recognition.

Evaluate your speech-to-text system with similarity measures such as word error rate (WER)

STT ?? ?? ??? ?? ????? ?? ???(CER), ?? ???(WER)? ???? Python ?? ???

Computes the MWER (minimum WER) Loss with CTC beam search. Knowledge distillation for CTC loss.

Evaluate results from ASR/Speech-to-Text quickly

Optimized loss based on cross-entropy (CE), like MWER (minimum WER) Loss with beam search and negative sampling strategy, Smoothed Max Pooling Loss.

Test-time adaptation for speech recognition model by single utterance. The official implementation of "Listen, Adapt, Better WER: Source-free Single-utterance Test-time Adaptation for Automatic Speech Recognition" paper.

The project,being part of Kagglex BIPOC Mentorship Program final project, aims to train two separate Hindi ASR models using the Facebook Wav2Vec2 (300M parameters) and OpenAI Whisper-Small models, respectively. The goal is to compare their performance, with a target WER of less than 13%, across various Hindi accents and dialects.

?? Easy-to-use Python package for lightning-fast Word Error Rate (WER) analysis

WhisperCPP INT8, INT4, INT5, INT6 quantization effect on model latency and WER experiment

Comparative analysis of the Moonshine and Faster-Whisper Tiny transcription models, focusing on latency and Word Error Rate (WER)

The automatic system that can extract PRAAT-like speech features from raw speech wav files, and also can get low WER (0.02) high quality transcriptions at the same time.

Go CLI wrapper around SCTK binaries for word error rate evaluation and error analysis for ASR systems.

A fine-tuned Whisper model (whisper-large-v2) for aviation audio transcription. WER < 5%.

Developed a Marathi speech-to-text application using the Hugging Face whisper ASR models. Trained the model with a custom audio dataset and fine-tuned it for optimized performance. Deployed the model on the Hugging Face Model Hub, achieving a WER of 0.74 for the base model.

This system is currently in prototype phase. This system can perform speech-to-text with WER at 0.4 approx.

Fine-tuned Whisper-base for STT (WER~10) and Roberta for NER (Precision~0.97) in Uzbek. Built using Common Voice and Uzbek NER datasets