Speech recognition software for Linux

In the late 1990s, a Linux version of ViaVoice, created by IBM, was made available to users for no charge. In 2002, the free software development kit (SDK) was removed by the developer.

In the early 2000s, there was a push to get a high-quality Linux native speech recognition engine developed. As a result, several projects dedicated to creating Linux speech recognition programs were begun, such as Mycroft, which is similar to Microsoft Cortana, but open-source.

It is essential to compile a speech corpus to produce acoustic models for speech recognition projects. VoxForge is a free speech corpus and acoustic model repository that was built to collect transcribed speech to be used in speech recognition projects. VoxForge accepts crowdsourced speech samples and corrections of recognized speech sequences. It is licensed under a GNU General Public License (GPL).

The first step is to begin recording an audio stream on a computer. The user has two main processing options:

• Discrete speech recognition (DSR) – processes information on a local machine entirely. This refers to self-contained systems in which all aspects of SR are performed entirely within the user's computer. This is becoming critical for protecting intellectual property (IP) and avoiding unwanted surveillance (2018).
• Remote or server-based SR – transmits an audio speech file to a remote server to convert the file into a text string file. Due to recent cloud storage schemes and data mining, this method more easily allows surveillance, theft of information, and inserting malware.

Remote recognition was formerly used by smartphones because they lacked sufficient performance, working memory, or storage to process speech recognition within the phone. These limits have largely been overcome although server-based SR on mobile devices remains universal.

Discrete speech recognition can be performed within a web browser and works well with supported browsers. Remote SR does not require installing software on a desktop computer or mobile device as it is mainly a server-based system with the inherent security issues noted above.

• Remote: The dictation service records an audio track of the user via a web browser.
• DSR: Some solutions work on a client only, without sending data to servers.

The following is a list of projects dedicated to implementing speech recognition in Linux, and major native solutions. These are not end-user applications. These are programming libraries that may be used to develop end-user applications.

• CMU Sphinx is a general term to describe a group of speech recognition systems developed at Carnegie Mellon University.
• HTK is the most famous and widely used speech recognition software before Kaldi.
• Julius is a high-performance, two-pass large vocabulary continuous speech recognition (LVCSR) decoder software for speech-related researchers and developers.
• Kaldi is a toolkit for speech recognition provided under the Apache licence.
• Mozilla DeepSpeech is developing an open-source Speech-To-Text engine based on Baidu's deep speech research paper.[1]

• VoxForge is a free speech corpus and acoustic model repository for open-source speech recognition engines.

• Janus Recognition Toolkit (JRTk) is a closed source speech recognition toolkit mainly targeted at Linux developed by the Interactive Systems Laboratories developed at Carnegie Mellon University and Karlsruhe Institute of Technology for which commercial and research licenses are available.[2]

It is possible to use programs such as Dragon NaturallySpeaking in Linux, by using Wine, though some problems may arise, depending on which version is used.[3]

It is also possible to use Windows speech recognition software under Linux. Using no-cost virtualization software, it is possible to run Windows and NaturallySpeaking under Linux. VMware Server or VirtualBox support copy and paste to/from a virtual machine, making dictated text easily transferable to/from the virtual machine.