Decibri: A Unified Technology Stack for Voice AI Applications
Decibri
Freemium Audio

Decibri

Decibri is a comprehensive technology stack designed for developing Voice AI applications, intelligent agents, and interactive devices in real time.

Go to Decibri
Categories
Audio
Pricing model
Freemium
Website
decibri.com

About

Decibri is a comprehensive technology stack designed for developing Voice AI applications, intelligent agents, and interactive devices in real time. It offers essential audio infrastructure features such as direct microphone recording, speaker playback, audio processing, speech activity detection, and streaming audio for downstream speech or agent services. Unlike using separate OS tools and language-specific audio layers, Decibri provides a unified foundation that works consistently across different platforms and programming languages.

Key features of Decibri include low-latency audio input/output with direct access to the OS audio system, and consistent audio chunking (100ms chunks by default). It also includes an Audio Conditioning Engine (ACE) that enhances audio quality before it is processed by speech models. This includes functions like voice activity detection (VAD), noise reduction, acoustic echo cancellation (AEC), automatic gain control, multichannel handling, and resampling. Recorded audio can also be processed using these conditioning features.

Decibri is flexible and can integrate with your preferred speech processing stack. It outputs raw PCM audio data (16-bit signed integer PCM by default, little-endian), allowing you to stream audio to local engines or cloud speech-to-text services. You can create fully offline processing pipelines or hybrid deployments while maintaining control over models, providers, and configurations. It comes with a pre-bundled Silero VAD v6.2 model that runs locally using ONNX Runtime.

Decibri supports Python, Node.js, and Rust, providing pre-built binaries and a command-line interface (CLI), including browser support for the Node.js package via conditional AudioWorklet exports. Installation is straightforward with a single command (e.g., pip install decibri), and it follows a freemium model with freely available components and commercial offerings.

Decibri outputs audio in 16-bit signed integer PCM format (little-endian by default), but a 32-bit float output format is also available for workflows that require it.

Decibri directly accesses the operating system’s audio layer through its Audio Host—WASAPI on Windows, CoreAudio on macOS, and ALSA on Linux—eliminating the need for external audio tools and intermediate encoding steps.

Pre-built binaries are provided for supported platforms, allowing pip and npm installations without the need for compilers, node-gyp, or post-install downloads. The Rust crate is compiled using cargo.

Node.js and browser usage are supported. When bundled for the browser (e.g., webpack or Vite), conditional exports provide the browser implementation (AudioWorklet), eliminating the need for a separate package.

Need content for a new website?AIGolova writes copy and creates images and video in one dashboard, with a single credit balance.
Try the Free demo
Freemium
Audio
Senzia

Senzia is a browser-based studio designed for creating AI-generated videos, images, and audio. You can work without switching between different chatbots, image platforms, and editors.

Freemium
Audio
VM Hunter

VM Hunter utilizes SpeechLLM to differentiate between live humans and answering machines, enhancing the reliability of call routing in outbound campaigns.