S

speech-to-text

Transcribes audio files to text locally with WhisperX.

Local audio-to-text web app. Upload an audio file (MP3/WAV/M4A), a background job transcribes it with WhisperX, then view and download the transcript as JSON, SRT, TXT, or Markdown.

Spring Boot + Thymeleaf frontend, Python worker under the hood.

Runs locally, no cloud.

A robust CLI that turns recordings and PDFs into clean, structured Markdown notes end-to-end.

Audio — chunks with configurable length and overlap, transcribes via a dedicated Whisper STT endpoint or cascades to OpenAI-compatible multimodal and Native Ollama backends.

PDF — text PDFs pass through pdftotext straight to refinement; scanned PDFs are auto-detected by character count, rendered to page images, and read by a vision model (or tesseract via an explicit --ocr escape, never silently).

LLM refinement — dedupes overlapping boundaries and formats the raw transcript into clean notes.

Resumable content-addressed cache — keyed by input hash with per-chunk/page signatures; a changed parameter invalidates only what depends on it.

Config resolves CLI > env var > settings file > defaults, including persistent chunk sizes. Python 3.12, built with uv, 147 tests at 100% coverage. Architecture is protocol-based and documented via TIPs (docs/).

Speech Note Linux app. Note taking, reading and translating with offline Speech to Text, Text to Speech and Machine translation

Audio capture service for the Oremi smart home ecosystem with wake word detection, speech transcription, and Home Assistant integration via MQTT.

https://demsking.gitlab.io/oremi-device/

Empowering seamless transcription with cutting-edge STT (Speech-to-Text) technology, revolutionizing interaction through accurate speech recognition.

https://demsking.gitlab.io/oremi-andika/

Phonic is an open-source AI-powered call interception and communication intelligence system that automatically screens incoming calls, blocks robocalls, records and transcribes voicemails, and maintains a complete call log. Built with modern telephony, machine learning, and speech-to-text technologies, it acts as an intelligent privacy layer between users and the phone network, continuously learning to improve spam detection and call handling. https://roxanneardary.com/phonic/

Self-hosted speech server in one Docker image. OpenAI-compatible /v1/audio/transcriptions and /v1/audio/speech across 12 ASR models (Whisper, Parakeet, Canary, Sherpa-ONNX, Vosk) and 3 TTS engines (Kokoro, Qwen3-TTS voice cloning, Chatterbox Turbo). Live WebSocket ASR, file staging, MCP built in. CPU + CUDA images.

A second brain for conversations you're already having: listens to your mic and system audio as two separate streams, transcribes both live, works out when the other side shut up, and hands you a reply draft before you panic. It drafts, never sends or speaks for you. Alpha — live capture is Linux/PipeWire only.

Self-hosted Telegram bot that transcribes voice/audio/video using Whisper (Bun + Python).

Vertical Rust libraries - public collection of reusable tools and runtimes (git, terminal, system, media, memory)

A PWA that provides a method to access chat like messaging functionality by using Rest API integrations from the Rhea Generative Framework.

The Rhea client app is component within the Rhea Generative Framework.

Originally intended as a way to demonstrate functionality found within the Rhea Generative Framework. The client app was designed to both demonstrate functionality and provide a foundation to build other components such as live chat (embeddable, etc.).

It has evolved over time to include additional functions for demonstration:

Persona management (role activation and management) Speech-to-text and text-to-speech (browser independent, both part of Rhea's Generative Framework server-side components and available for local hosting) STT captures audio for x seconds, transcribes and offers to either continue transcribing, send as a message, or manually edit.

Sussurro - dettatura vocale locale in italiano per Windows, via Whisper su GPU. Zero cloud, zero costi ricorrenti.

Drop a file, get a transcript, nobody else ever hears it. Local Whisper + pyannote transcription with speaker labels, on your own machine, Windows and Linux.

Give your AI agent a voice. Local speech stack for Apple Silicon: TTS, ASR, forced alignment, voices, daemon, and MCP bridge.

System-wide voice-to-text for macOS. Hold a hotkey, speak, text appears at your cursor. Supports OpenAI, Gemini, Apple Speech & local models. No subscription — bring your own API keys.

Transcripty is a privacy-first, locally run transcription software using locally running AI-models for speech to text analysis. The software has light editing capability for splitting and merging text blocks and assigning speakers. The transcription result can be exported to a text doc for use in other applications.

MacWhisperer is a macOS desktop GUI for OpenAI Whisper that makes local speech transcription and English translation easy, with auto model loading, live recording, input device controls, and file-based workflows.

The project provides an end-to-end solution for converting speech to text and analyzing content through semantic segmentation and summarization. It is especially useful for processing lectures, interviews, podcasts and other long audio content.launching the speech analyzer