Introduction

Picovoice is the developer-first platform for building voice AI and LLM-powered products on your terms. You can use Picovoice for keyword spotting, voice commands, voice user interfaces (VUI), phonetic search, automatic speech recognition (ASR), speech-to-text (STT), voice activity detection (VAD), noise suppression, speech enhancement, speaker diarization, speaker recognition, text-to-speech (TTS), and LLM intelligence.

Free Plan

Start building for free. No credit card is required!

Platform Features

• Accurate: Resilient to noise and reverberation. Picovoice outperforms cloud-based alternatives and backs its claims with open-source benchmarks:
  • LLM Compression Benchmark
  • Wake Word Benchmark (KWS & hotword)
  • Speech-to-Intent Benchmark (VUI & NLU)
  • Speech-to-Text Benchmark (ASR & STT)
  • Text-to-Speech Latency Benchmark (Speech Synthesis)
  • Voice Activity Detection Benchmark (VAD)
  • Noise Suppression Benchmark (Speech Enhancement)
  • Speaker Diarization Benchmark (Diarization)
  • Speaker Recognition Benchmark (Speaker Recognition)
  • Speech-to-Index Benchmark (phonetic search)
• Private & Secure: Voice recognition is entirely offline. Intrinsically private; HIPAA and GDPR-compliant.
• Zero-Latency: Edge-first architecture eliminates unpredictable network delay.
• Self-Service: Design, train, and test voice interfaces instantly in your browser, using Picovoice Console.
• Cross-Platform: Design once, deploy anywhere. Build using familiar languages and frameworks.

Flow

Design

Picovoice Console is the cloud-based platform for training custom speech models, all within your web browser. No machine learning skills are required. Describe what you need with text and export trained models. Picovoice Console supports training branded wake words, context-aware voice commands, and custom ASR models. You can find ready-to-use compressed Large Language Models on Picovoice Console, and download the best fit for your application.

Develop

Exported models from Picovoice Console can run on Picovoice SDKs. Picovoice SDKs run on many platforms and support most frameworks. The Picovoice Console and SDKs enable you to design, build and iterate fast.

Deploy

Start for free by taking advantage of Picovoice Free Plan, even commercially. Deploy without having to maintain complex cloud infrastructure. Avoid hefty API-based pricing, limitations, and control imposed by big tech.

Getting Started

LLM Inference

Picovoice picoLLM Inference is the only cross-platform local LLM inference engine that runs X-bit quantized LLMs across Linux, macOS, Windows, Android, iOS, Chrome, Safari, Edge, Firefox, Raspberry Pi, or other embedded platforms, supporting both CPU and GPU, offering LLM Intelligence without sacrificing performance or privacy.

Speech-to-Text

Picovoice Leopard Speech-to-Text is the only on-device ASR engine that matches and exceeds cloud-level accuracy with minimal resource requirements. You can use Picovoice Console to train models that understand the custom vocabulary that matters to your use case.

Streaming Speech-to-Text

Picovoice Cheetah Streaming Speech-to-Text is the only on-device real-time ASR engine that matches and exceeds cloud-level accuracy, enabling real-time transcriptions with fast and guaranteed response time.

Streaming Text-to-Speech

Picovoice Orca Streaming Text-to-Speech offers guaranteed response time by eliminating network latency, enabling human-like fast interactions. Orca Text-to-Speech can read streaming LLM responses as they emerge for an even faster response time.

Noise Suppression

Picovoice Koala Noise Suppression is the only high-quality and cross-platform on-device noise suppression engine.

Speaker Recognition

Picovoice Eagle Speaker Recognition is the only enterprise-grade and freely-available speaker recognition engine.

Speaker Diarization

Picovoice Falcon Speaker Diarization is the most efficient and only cross-platform Speaker Diarization software that works with any Speech-to-Text engine.

Wake Word

Picovoice Porcupine Wake Word is the wake word detection engine. It detects utterances of given wake phrases. You can train custom wake words using Picovoice Console and then run the exported wake word models on the Picovoice SDK.

Voice Commands

Picovoice Rhino Speech-to-Intent engine directly infers users' intent from spoken commands within a given domain of interest (context). You can design and train custom contexts for your product using Picovoice Console.

Voice Activity Detection

Picovoice Cobra Voice Activity Detector engine detects the presence of speech in a stream of audio in real time.

Phonetic Search

Picovoice Octopus Speech-to-Index directly searches speech data without relying on a text representation. It outperforms cloud-based speech-to-text engines on search tasks by removing out-of-vocabulary and competing hypothesis problems.

Technical Support

Forever-Free Plan account owners can report bugs by creating a GitHub issue under the relevant repository and purchase the optional Support Add-on for dedicated support. Developer and Enterprise Plan account owners can reach out to their Picovoice contacts.

Commercial Inquiries

Please refer to our pricing page to learn about our business model.