Nomily-Ai/nomily-app

BYOK voice recording companion with ASR transcription, speaker diarization, AI summaries, and local ASR endpoint support.

What it solves

Nomily AI provides a privacy-focused workflow for converting voice recordings into structured text. It eliminates the need for a centralized, hosted transcription service by allowing users to bring their own API keys (BYOK) for transcription and summarization, ensuring the user maintains control over their chosen AI providers.

How it works

The system acts as a native client that manages audio recordings from various sources (V05E BLE recorders, iPhones, and Apple Watches). It then routes this audio to a user-configured ASR (Automatic Speech Recognition) provider—such as Azure Speech or a local faster-whisper endpoint—for transcription and speaker diarization. Finally, the resulting transcript is sent to a selected LLM provider (including OpenAI, Anthropic, Gemini, OpenRouter, or Ollama) to generate summaries, titles, and translations.

Who it’s for

It is designed for individuals who record audio on the go and want a native mobile experience (iOS/Android) for transcription and AI-powered synthesis without relying on a proprietary cloud account.

Highlights

  • Bring Your Own Key (BYOK): Users configure their own API keys for both transcription and LLM services.
  • Multimodal Input: Supports V05E BLE recorders, iPhone imports, and a dedicated Apple Watch companion app.
  • Flexible ASR: Supports Azure Speech for speaker diarization (up to 10 speakers) or local ASR endpoints.
  • Broad LLM Support: Integrates with major providers like OpenAI, Anthropic, Gemini, and local options via Ollama.

Related

  • Project
    Azure-Samples/Cognitive-Speech-TTS

    A Microsoft‑maintained repo of SDK and REST code samples showing how to use Azure Cognitive Services Speech (Text‑to‑Speech), with links to customer stories, blog posts, and research. Ideal for developers who need ready‑to‑run examples for realistic, multilingual synthetic voices, custom neural voices, and related Azure Speech features.

  • Project
    QwenAudio/Fun-Audio-Chat

    Fun-Audio-Chat is a Large Audio Language Model designed for natural, low-latency voice interactions, utilizing dual-resolution speech representations to reduce compute costs.

  • Project
    TensorSpeech/TensorFlowASR

    A TensorFlow 2 framework for Automatic Speech Recognition that implements multiple SOTA architectures and supports conversion to TFLite for efficient deployment.

  • Project
    ronibandini/reggaetonBeGone

    A Raspberry Pi‑3 gadget that runs an Edge Impulse audio‑classification model locally to detect reggaeton. When confidence exceeds 95 %, it triggers a Bluetooth routine that disables a configured speaker. Includes OLED feedback, a start button, logging, and full hardware‑software instructions.