From Things and Stuff Wiki
Jump to navigation Jump to search


See also Automation, Audio#Transcription, Synth vox

  • https://github.com/facebookresearch/seamless_communication - our foundational all-in-one Massively Multilingual and Multimodal Machine Translation model delivering high-quality translation for speech and text in nearly 100 languages. SeamlessM4T models support the tasks of: Speech-to-speech translation (S2ST), Speech-to-text translation (S2TT), Text-to-speech translation (T2ST), Text-to-text translation (T2TT), Automatic speech recognition (ASR)



  • ESPS - Entropic Signal Processing System, is a package of UNIX-like commands and programming libraries for speech signal processing. As a commercial product of Entropic Research Laboratory, Inc, it became extremely widely used in phonetics and speech technology research laboratories in the 1990's, in view of the wide range of functions it offered, such as get_f0 (for fundamental frequency estimation), formant (for formant frequency measurement), the xwaves graphical user interface, and many other commands and utilities. Following the acquisition of Entropic by Microsoft in 1999, Microsoft and AT&T licensed ESPS to the Centre for Speech Technology at KTH, Sweden, so that a final legacy version of the ESPS source code could continue to be made available to speech researchers. At KTH, code from the ESPS library (such as get_f0) was incorporated by Kåre Sjölander and Jonas Beskow into the Wavesurfer speech analysis tool. This is a very good alternative way to use many ESPS functions if you want a graphical user interface rather than scripting.

Speech Research Tools

  • https://sourceforge.net/projects/speechresearch - Software for speech research. It includes programs and libraries for signal processing, along with general purpose scientific libraries. Most of the code is in Python, with C/C++ supporting code. Also, contains code releases corresponding to publishe

HTK Speech Recognition Toolkit

  • HTK Speech Recognition Toolkit - a portable toolkit for building and manipulating hidden Markov models. HTK is primarily used for speech recognition research although it has been used for numerous other applications including research into speech synthesis, character recognition and DNA sequencing. HTK is in use at hundreds of sites worldwide.HTK consists of a set of library modules and tools available in C source form. The tools provide sophisticated facilities for speech analysis, HMM training, testing and results analysis. The software supports HMMs using both continuous density mixture Gaussians and discrete distributions and can be used to build complex HMM systems. The HTK release contains extensive documentation and examples.


  • speechrate - software for the analysis of speech. Below you will find a script that automatically detects syllable nuclei in order to measure speech rate without the need of a transcription. Peaks in intensity (dB) that are preceded and followed by dips in intensity are considered as potential syllable nuclei. The script subsequently discards peaks that are not voiced. On this page you find an example of how the script works.


  • Higgins Annotation Tool - can be used to transcribe and annotate speech with one or more audio tracks (such as dialogue). Windows.



  • https://github.com/wjassim/WARP-Q - code is to run the WARP-Q speech quality metric. WARP-Q (Quality Prediction For Generative Neural Speech Codecs) is an objective, full-reference metric for perceived speech quality. It uses a subsequence dynamic time warping (SDTW) algorithm as a similarity between a reference (original) and a test (degraded) speech signal to produce a raw quality score. It is designed to predict quality scores for speech signals processed by low bit rate speech coders.




  • https://en.wikipedia.org/wiki/Speech_recognition - an interdisciplinary subfield of computational linguistics that develops methodologies and technologies that enables the recognition and translation of spoken language into text by computers. It is also known as automatic speech recognition (ASR), computer speech recognition or speech to text (STT). It incorporates knowledge and research in the linguistics, computer science, and electrical engineering fields.




  • Audiogrep - transcribes audio files and then creates "audio supercuts" based on search phrases. It uses CMU Pocketsphinx for speech-to-text and pydub to stitch things together. [3]

Speech Signal Processing Toolkit

  • Speech Signal Processing Toolkit (SPTK) - a suite of speech signal processing tools for UNIX environments, e.g., LPC analysis, PARCOR analysis, LSP analysis, PARCOR synthesis filter, LSP synthesis filter, vector quantization techniques, and other extended versions of them. This software is released under the Modified BSD license. SPTK was developed and has been used in the research group of Prof. Satoshi Imai (he has retired) and Prof. Takao Kobayashi (currently he is with Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology) at P&I laboratory, Tokyo Institute of Technology. A sub-set of tools was chosen and arranged for distribution by Prof. Keiichi Tokuda (currently he is with Department of Computer Science and Engineering, Nagoya Institute of Technology) as a coordinator in cooperation and other collaborates (see "Acknowledgments" and "Who we are" in README).The original source codes have been written by many people who took part in activities of the research group. The most original source codes of this distribution were written by Takao Kobayashi (graph, data processing, FFT, sampling rate conversion, etc.), Keiichi Tokuda (speech analysis, speech synthesis, etc.), and Kazuhito Koishida (LSP, vector quantization, etc.).



to sort



  • https://github.com/k2-fsa/k2 - The vision of k2 is to be able to seamlessly integrate Finite State Automaton (FSA, and Finite State Transducer (FST) algorithms into autograd-based machine learning toolkits like PyTorch and TensorFlow. For speech recognition applications, this should make it easy to interpolate and combine various training objectives such as cross-entropy, CTC and MMI and to jointly optimize a speech recognition system with multiple decoding passes including lattice rescoring and confidence estimation. We hope k2 will have many other applications as well.


  • https://github.com/lhotse-speech/lhotse - a Python library aiming to make speech and audio data preparation flexible and accessible to a wider community. Alongside k2, it is a part of the next generation Kaldi speech processing library.



  • RWTH ASR - The RWTH Aachen University Speech Recognition System, a software package containing a speech recognition decoder together with tools for the development of acoustic models, for use in speech recognition systems. It has been developed by the Human Language Technology and Pattern Recognition Group at the RWTH Aachen University since 2001. Speech recognition systems developed using this framework have been applied successfully in several international research projects and corresponding evaluations.


  • https://github.com/espnet/espnet - an end-to-end speech processing toolkit, mainly focuses on end-to-end speech recognition and end-to-end text-to-speech. ESPnet uses chainer and pytorch as a main deep learning engine, and also follows Kaldi style data processing, feature extraction/format, and recipes to provide a complete setup for speech recognition and other speech processing experiments.


  • RETURNN - extensible training framework for universal recurrent neural networks, is a Theano/TensorFlow-based implementation of modern recurrent neural network architectures. It is optimized for fast and reliable training of recurrent neural networks in a multi-GPU environment.




  • voice2json - a collection of command-line tools for offline speech/intent recognition on Linux. It is free, open source, and supports 16 languages.


  • https://github.com/fredvs/sak - Speecher Assistive Kit. With sak, your application becomes assistive directly, without changing anything in your code. sak uses the PortAudio and eSpeak Open Source Libraries.




  • https://github.com/Aria-K-Alethia/ExVo - Official implementation of the paper "Exploring the Effectiveness of Self-supervised Learning and Classifier Chains in Emotion Recognition of Nonverbal Vocalizations" accepted by the Few-Shot track of the ICML Expressive Vocalizations (ExVo) Competition 2022.

Neural net


to sort

  • Sequence Modeling with CTC - A visual guide to Connectionist Temporal Classification, an algorithm used to train deep neural networks in speech recognition, handwriting recognition and other sequence problems. [6]

NICO toolkit

  • NICO toolkit - mainly intended for, and originally developed for speech recognition applications, a general purpose toolkit for constructing artificial neural networks and training with the back-propagation learning algorithm. The network topology is very flexible. Units are organized in groups and the group is a hierarchical structure, so groups can have sub-groups or other objects as members. This makes it easy to specify multi-layer networks with arbitrary connection structure and to build modular networks.


  • https://github.com/mozilla/DeepSpeech - an open source Speech-To-Text engine, using a model trained by machine learning techniques based on Baidu's Deep Speech research paper. Project DeepSpeech uses Google's TensorFlow to make the implementation easier.



  • https://github.com/bytedance/neurst - aims at easily building and training end-to-end speech translation, which has the careful design for extensibility and scalability. We believe this design can make it easier for NLP researchers to get started. In addition, NeurST allows researchers to train custom models for translation, summarization and so on.







  • https://github.com/santi-pdp/pase - This repository contains the official implementations of PASE and PASE+. These are speech waveform encoders trained in a self-supervised manner with the so called worker/minion framework. A PASE model can be used as a speech feature extractor or to pre-train an encoder for our desired end-task, like speech classification such as in ASR, seaker recognition, or emotion recognition, or speech generation such as in voice conversion or TTS.






  • https://github.com/Aria-K-Alethia/ExVo - Official implementation of the paper "Exploring the Effectiveness of Self-supervised Learning and Classifier Chains in Emotion Recognition of Nonverbal Vocalizations" accepted by the Few-Shot track of the ICML Expressive Vocalizations (ExVo) Competition 2022. Abstract: We present an emotion recognition system for nonverbal vocalizations (NVs) submitted to the ExVo Few-Shot track of the ICML Expressive Vocalizations Competition 2022. The proposed method uses self-supervised learning (SSL) models to extract features from NVs and uses a classifier chain to model the label dependency between emotions. Experimental results demonstrate that the proposed method can significantly improve the performance of this task compared to several baseline methods. Our proposed method obtained a mean concordance correlation coefficient (CCC) of in the validation set and in the test set, while the best baseline method only obtained in the validation set.


  • https://github.com/Vaibhavs10/insanely-fast-whisper - Powered by Transformers, Optimum & flash-attn. TL;DR - Transcribe 150 minutes (2.5 hours, of audio in less than 98 seconds - with OpenAI's Whisper Large v3. Blazingly fast transcription is now a reality!


See also Speech


  • Transcribe! - an assistant for people who want to work out a piece of music from a recording, in order to write it out, or play it themselves, or both. It doesn't do the transcribing for you, but it is essentially a specialised player program which is optimised for the purpose of transcription. It has many transcription-specific features not found on conventional music players. It is also used by many people for play-along practice. It can change pitch and speed instantly, and you can store and recall any number of named loops. So you can practice in all keys, and you can speed up as well as slow down. There is some advice about play-along practice in Transcribe!'s help, under the heading "Various Topics". And it is also used for speech transcription. With its support for foot pedals and its superior slowed-down sound quality, it is an excellent choice for this purpose. There is some advice about speech transcription in Transcribe!'s help, under the heading "Various Topics".


  • https://github.com/google/live-transcribe-speech-engine - an Android application that provides real-time captioning for people who are deaf or hard of hearing. This repository contains the Android client libraries for communicating with Google's Cloud Speech API that are used in Live Transcribe.


  • https://github.com/botbahlul/Live-Subtitle - ANDROID APP that can RECOGNIZE VLC LIVE AUDIO/VIDEO STREAMING (using free Android Developers Speech Recognition API, then TRANSLATE (using ANDROID MLKIT TRANSLATE API) and display it as LIVE CAPTION / LIVE SUBTITLE


  • https://github.com/magenta/mt3 - a multi-instrument automatic music transcription model that uses the T5X framework.This is not an officially supported Google product.





Video Subtitle Generator


  • https://github.com/uhh-lt/bbb-live-subtitles - a plugin for automatic subtitling in BigBlueButton (BBB), an open source web conferencing system. bbb-live-subtitles will run real time automatic speech recognition (ASR, and will generate subtitle captions on-the-fly. No cloud services are used for ASR, instead we use our own speech recognition models that can be run locally. This ensures that no privacy issues arise. There are pre-trained German and English models available and ready to use (English needs PyKaldi > 0.2.0 and Python 3.8).


  • https://github.com/cxhawk/liveSubs - Wanna add subtitles and lower third easily in your live stream? This tool is great to be used with a video switcher like the ATEM Minis. Make sure this source is configured to be chroma keyed with the background color in settings.


Voice control

  • YouTube: "Voice Driven Development: Who needs a keyboard anyway?" by Emily Shea - As a programmer, being unable to use a keyboard can seem like a career-ending limitation. I was facing that reality a year ago due to Repetitive Strain Injury, when every productive day was a step back for my health. In a futile, last-ditch effort at saving my career, I turned to speech recognition. Unexpectedly, not only was I able to return to pain-free productivity, but I found voice-driven development to be enjoyable, efficient, and a source of renewed excitement for programming. Speech recognition isn't exactly known in software development circles as a workable approach to programming, but it's better than you (probably) think! Far from imposing a tedious workflow with a specialized set of commands, speech can enable flexibility and optimizations that are difficult to achieve with a keyboard, and does not get in the way when adapting to new technologies, tools, or languages. This practical, demo-driven talk presents the approach and tools that have allowed me to continue on with my career. I'll present real world experience with speech-related topics like disambiguating homophones on the fly and teaching the voice engine new technical vocabulary. Beyond technical aspects, I will share my experience guiding a team to support this working style, and challenges to adding a microphone into an open office environment. Speech not only removes the compromise between health and career when an injury is present, but can also augment and improve the workflow for a wider audience.



  • Simon - an open source speech recognition program that can replace your mouse and keyboard. The system is designed to be as flexible as possible and will work with any language or dialect.


  • CVoiceControl - a speech recognition system that enables a user to connect spoken commands to unix commands. It automagically detects speech input from a microphone, performs recognition on this input and - in case of successful recognition - executes the associated unix command.

Perlbox Voice

  • Perlbox Voice - an voice enabled application to bring your desktop under your command. With a single word, you can start your web browser, your favorite editor or whatever you want. Perlbox Voice now has desktop plugins, which will allow you to control desktop functions. You can switch virtual desktops, invoke the desktop menu, switch wallpaper or lock the screen. This is command, and it's right on the tip of your tongue! Start with the tutorial, which has a lot of screenshots.


  • Xvoice - enables continuous speech dictation and speech control of most X applications. To convert users' speech into text it uses the IBM ViaVoice speech recognition engine, which is distributed separately (see below).When in dictation mode Xvoice passes this text directly to the currently focused X application. When in command mode, Xvoice matches the speech with predefined, user-modifieable, key sequences or commands. For instance "list" would match "ls -l" when commanding the console, so that when the user says "list" "ls -l" will be sent to the console as if the user had typed it.


  • Cursorless - Voice coding at the speed of thought. Cursorless is a spoken language for structural code editing, enabling developers to code by voice at speeds not possible with a keyboard. Cursorless decorates every token on the screen and defines a spoken language for rapid, high-level semantic manipulation of structured text.


  • Numen - voice control letting you have full control of a Linux machine without needing to type. It empowers people who otherwise couldn't use their computers, and helps more avoid hand strain. Numen is free (libre) software and runs locally. [10]





See also Effects#Noise reduction


  • https://github.com/nay0648/unified2021 - A UNIFIED SPEECH ENHANCEMENT FRONT-END FOR ONLINE DEREVERBERATION, ACOUSTIC ECHO CANCELLATION, AND SOURCE SEPARATION. Dereverberation (DR), acoustic echo cancellation (AEC), and blind source separation (BSS) are the three most important submodules in speech enhancement front-end. In traditional systems, the three submodules work independently in a sequential manner, each submodule has its own signal model, objective function, and optimization policy. Although this architecture has high flexibility, the speech enhancement performance is restricted, since each submodule's optimum cannot guarantee the entire system's global optimum. In this paper, a unified signal model is derived to combine DR, AEC, and BSS together, and the online auxiliary-function based independent component/vector analysis (Aux-ICA/IVA) technique is used to solve the problem. The proposed approach has unified objective function and optimization policy, the performance improvement is verified by simulated experiments.


DNN-based source separation


  • https://github.com/rrbluke/BSSD - contains python/tensorflow code to reproduce the experiments presented in our paper Blind Speech Separation and Dereverberation using Neural Beamforming.


Ultimate Vocal Remover GUI

  • https://github.com/Anjok07/ultimatevocalremovergui - This application uses state-of-the-art source separation models to remove vocals from audio files. UVR's core developers trained all of the models provided in this package (except for the Demucs v3 and v4 4-stem models).

Resemble Enhance

  • https://github.com/resemble-ai/resemble-enhance - an AI-powered tool that aims to improve the overall quality of speech by performing denoising and enhancement. It consists of two modules: a denoiser, which separates speech from a noisy audio, and an enhancer, which further boosts the perceptual audio quality by restoring audio distortions and extending the audio bandwidth. The two models are trained on high-quality 44.1kHz speech data that guarantees the enhancement of your speech with high quality.