Kaldi alternatives for speech recognition
Kaldi is a free, open-source C++ toolkit for speech-recognition researchers. It runs offline and in real time, but Windows examples do not work and its documentation targets experts.
These alternatives shift toward speech APIs, broader language coverage, or simpler desktop use. Choose an open-source toolkit for building systems, an API for app integration, or a ready-to-use transcription app.
S
Speechmatics: speech APIs for live and recorded audioAPIOpenFreeSpeechmatics provides APIs for speech-to-text, voice agents and text-to-speech, handling recorded files or live audio. It runs on Linux, macOS, Windows, the web and self-hosted setups, while supporting more than 55 languages and mid-sentence language switching. Kaldi is a free C++ toolkit with offline and real-time recognition, downloadable models and research recipes, but its Windows support is not actively maintained.
Trade-off: Speechmatics is freemium, with paid plans from $0.13; Kaldi is free and open source. Suits: Teams needing speech APIs, live audio, or multilingual conversations.
E
ESPnet: free open-source speech toolkitAPIOpenFreeESPnet is a free, open-source option for speech work, with API, Linux, macOS, self-hosted and Windows platforms and permission for commercial use. It may suit teams that want a research toolkit without Kaldi's Windows maintenance limits. Kaldi offers Android and WebAssembly support, real-time and offline recognition, OpenFst integration, downloadable models and established C++ recipes.
Trade-off: ESPnet lists no voice cloning and fewer stated Kaldi-specific models and integrations. Suits: Researchers and builders wanting an open-source toolkit with Windows support.
SSpeechRecognition: Python library with online and offline enginesAPIOpenFree SpeechRecognition is a free Python library that connects to several recognition engines and APIs, including CMU Sphinx, Google, Azure, Vosk and Whisper. Its README identifies offline options, giving app developers a simpler way to switch engines. Kaldi is a C++ toolkit with its own recipes, models, real-time and offline operation, plus Android and WebAssembly paths.
Trade-off: It depends on supported engines; Kaldi provides its own toolkit, models and training recipes. Suits: Python developers who need a choice of online or offline engines.
S
Superwhisper: formatted speech-to-text across applicationsAPIOpenFreeSuperwhisper turns speech into polished, formatted text across applications instead of providing a research toolkit. It runs on Android, iOS, Linux, macOS and Windows and offers an API, speaker identification, a free plan and a trial; paid plans start at $8.49 per month. Kaldi is a free C++ toolkit with offline and real-time recognition, downloadable models and research recipes.
Trade-off: Superwhisper is not open source and paid plans start at $8.49 per month. Suits: People who want ready-to-use dictation across desktop and mobile apps.
A
AssemblyAI: speech API with timestamps and captionsAPIOpenFreeAssemblyAI provides an API for speech recognition with 99 languages, speaker identification, timestamps and SRT or VTT exports. It offers web, API and self-hosted platforms, plus a free plan, with pricing from $0.15. Kaldi is free and open source, runs offline and in real time, and includes downloadable models, but requires source installation and expert-oriented setup.
Trade-off: AssemblyAI charges from $0.15, while Kaldi is free; Kaldi also supports offline operation. Suits: App teams needing multilingual transcription, timestamps and caption files.
GCGoogle Cloud Speech-to-Text: transcription API for applicationsAPIOpenFree Google Cloud Speech-to-Text converts audio into text and provides APIs for integrating recognition into applications. It supports API, web and self-hosted platforms, speaker identification, timestamps, and VTT or SRT exports, with a free plan and pricing from $0.00. Kaldi is a free, open-source toolkit for researchers that can run offline on a device or Unix-like machine.
Trade-off: Google Cloud is a paid service; Kaldi is free, open source and can run offline. Suits: Developers integrating transcription into applications through an API.
AT
Amazon Transcribe: transcription with 113 languagesAPIOpenFreeAmazon Transcribe offers API and web access, 113 supported languages, speaker identification, timestamps and JSON, WebVTT or SRT exports. It has a free plan and pricing from $0.01, but no free trial. Kaldi is free and open source, supports offline and real-time recognition, and can run self-hosted across Linux and macOS with downloadable models.
Trade-off: Amazon Transcribe offers API and web access and is paid from $0.01; Kaldi supports offline self-hosting at no cost. Suits: Teams needing broad language coverage and caption or JSON exports.
More near it on Speech Recognition Software
- 16 N
Nottanotta.ai - 18 OA
Otter.aiotter.ai - 19 S
Sonioxsoniox.com - 20 VIVoice Indictanote.co
Voice In is a browser extension for speech-to-text dictation in web text fields, typing recognized speech at the cursor in real time.
- 13 TTrinttrint.com
Trint is an AI-powered SaaS platform for transcription, editorial tools, collaboration, and content publishing.
- 21 NR
NVIDIA Rivadeveloper.nvidia.comNVIDIA Riva is a GPU-accelerated collection of multilingual speech and translation microservices for real-time conversational AI.
- 22 TVTalon Voicetalonvoice.com
Talon provides hands-free computer input for writing, coding, browsing and gaming.
- 23 CSCMU Sphinxcmusphinx.github.io
CMU Sphinx is an open-source speech recognition toolkit for building speech applications.
- 10 IW
IBM Watson Speech to Textibm.comThe service uses speech recognition APIs to transcribe spoken audio in multiple languages and formats.
- 24 YSYandex SpeechKityandex.cloud
- 9 RA
Rev AIrev.aiRev AI provides APIs for asynchronous and real-time speech-to-text transcription plus speech insights.
- 25 M
MacWhispermacwhisper.com - 8 AA
Azure AI Speechazure.microsoft.com - 27 DPDragon Professionaldragon.nuance.com
Questions
Is there a free alternative to Kaldi?
ESPnet and SpeechRecognition are free options. ESPnet is open source with commercial use, while SpeechRecognition is a free Python library supporting several engines.
Which alternative works on Windows?
ESPnet, SpeechRecognition, Superwhisper and Speechmatics list Windows support. Kaldi can compile on Windows, but its example scripts do not work there and compatibility is not actively maintained.
Which tools offer an API?
Superwhisper, AssemblyAI, Gladia, Google Cloud Speech-to-Text, Amazon Transcribe, Rev AI, Speechmatics, ESPnet, Soniox and SpeechRecognition list API access.
Can I run these alternatives offline?
SpeechRecognition lists CMU Sphinx, Vosk and Whisper as offline options. Kaldi also supports offline recognition and self-hosted deployment.
Which alternatives identify speakers?
Superwhisper, AssemblyAI, Gladia, Google Cloud Speech-to-Text, Amazon Transcribe and Rev AI list speaker identification.