# Automatic Speech Recognition

parakeet-tdt-0.6b-v2 is a 600 million parameter automatic speech recognition (ASR) model designed to achieve high-quality English transcription with accurate timestamp prediction and automatic punctuation and capitalization support. The model is based on the FastConformer architecture, capable of efficiently processing audio clips up to 24 minutes long, making it suitable for developers, researchers, and various industry applications.

PengChengStarling is an open-source toolkit focused on multilingual automatic speech recognition (ASR), developed based on the icefall project. It supports the entire ASR process, including data processing, model training, inference, fine-tuning, and deployment. By optimizing parameter configurations and integrating language identifiers into the RNN-Transducer architecture, it significantly enhances the performance of multilingual ASR systems. Its main advantages include efficient multilingual support, a flexible configuration design, and robust inference performance. The models in PengChengStarling perform exceptionally well across various languages, require relatively small model sizes, and offer extremely fast inference speeds, making it suitable for scenarios that demand efficient speech recognition.

BetterWhisperX is an improved automatic speech recognition model based on WhisperX, offering fast speech-to-text services with word-level timestamps and speaker identification features. This tool is vital for researchers and developers handling large volumes of audio data, as it significantly enhances the efficiency and accuracy of speech data processing. The product is built on OpenAI's Whisper model with further optimizations and improvements. Currently, the project is free and open-source, aiming to provide the developer community with more efficient and accurate speech recognition tools.

Whisper-NER is an innovative model that allows for simultaneous speech transcription and entity recognition. This model supports open-type Named Entity Recognition (NER) and can identify a diverse and evolving set of entities. Whisper-NER is designed as a robust foundational model for automatic speech recognition (ASR) and NER downstream tasks and can be fine-tuned on specific datasets to enhance performance.

WhisperNER is a unified model that combines Automatic Speech Recognition (ASR) and Named Entity Recognition (NER), equipped with zero-shot capabilities. This model is designed as a robust foundational model for downstream tasks of ASR with NER and can be fine-tuned on specific datasets to enhance performance. The significance of WhisperNER lies in its ability to simultaneously handle speech recognition and entity recognition tasks, improving efficiency and accuracy, especially in multilingual and cross-domain scenarios.

Moonshine is a suite of speech-to-text models optimized for resource-constrained devices, making it ideal for real-time, on-device applications such as live transcription and voice command recognition. It outperforms the OpenAI Whisper model of the same size in word error rate (WER) on test datasets used in the OpenASR leaderboard maintained by HuggingFace. Additionally, Moonshine's computational requirements vary with the length of the input audio, allowing for quicker processing of shorter audio compared to the Whisper model, which processes everything in 30-second chunks. Moonshine processes 10-second audio segments at a speed five times faster than Whisper while maintaining the same or better WER.

CrisperWhisper is an advanced variant of OpenAI's Whisper model, specifically designed for fast, accurate, verbatim speech recognition, providing precise word-level timestamps. Unlike the original Whisper model, CrisperWhisper aims to transcribe every spoken word, including filler words, pauses, stutters, and false starts. This model ranks first in word-level datasets such as TED and AMI, and has been accepted at INTERSPEECH 2024.

seed-tts-eval is a testing dataset for evaluating a model's zero-shot speech generation capability. It provides an objective evaluation test set across diverse domains, containing samples extracted from both English and Mandarin public language repositories. This dataset is used to measure the model's performance across various objective metrics. It utilizes 1000 samples from the Common Voice dataset and 2000 samples from the DiDiSpeech-2 dataset.

WhisperKit is a tool for compressing and optimizing automatic speech recognition (ASR) models. It allows for model compression and optimization, providing detailed performance evaluation data. WhisperKit also offers quality assurance certifications for different datasets and model formats, and supports local reproducibility of test results.