Embed private voice AI into any product instantly


company logo

Picovoice provides core technology to embed private voice AI into any device. Picovoice's SDK makes it possible to build products that can be activated, controlled, and conversed with using voice without cloud connection. Our software runs on embedded processors, Raspberry Pi, Android, iOS, watchOS, Linux, Mac, Windows, and even web browsers.

private offline on-device

Private

Runs locally without an internet connection. Nothing is sent to the cloud to fully protect user's data.

efficient embedded IoT

Efficient

Uses proprietary deep learning technology that enables cutting-edge models to run on commodity (embedded) hardware efficiently.

accurate robust

Accurate

Resilient to noise, reverberation, and works across a variety of accents. Works everywhere for anyone.

plug and play customizable works out-of-the-box

Plug and Play

Enables customizations within seconds. Reduces time to market while promoting user's product, brand, and identity.

cross-platform Linux Mac Windows iOS watchOS Android Raspberry Pi ARM Cortex-A ARM Cortex-M Tensilica HiFi

Cross-Platform

Runs across all major platforms. Linux, Mac, Windows, Android, iOS, watchOS, Raspberry Pi, ARM Cortex-A, ARM Cortex-M, and a growing number of embedded processors are supported.

open benchmark metric

Open Benchmarks

Picovoice empowers users to evaluate its technology independently. Facilitates data-driven decision making.

Picovoice's wake word detection library enables building products that can be activated and controlled using voice. It makes it possible to activate the device similar to “Alexa” or “OK Google” but using your hotword (wake word) of choice to promote your product, brand, and identity. Additionally, the library allows the user to control the device via a configurable set of voice commands.

zero lead time instant zero wait time

Zero Lead Time

Uses proprietary AI algorithms to build models for any custom wake word/command within seconds (instead of weeks). Removes the hassle of time-consuming and costly data gathering phase.

scalable multiple voice commands

Multiple Phrase Detection

Can detect many hotwords (wake up words and commands) concurrently with no additional CPU or memory footprint.

lightweight fast embedded IoT

Lightweight

Natural fit for IoT. Can run with as low as 240 KB of memory and 3.4% CPU usage on Raspberry Pi 3.

accurate optimized

Highly-Accurate

Outperforms existing solutions with high margins in clean and noisy environments. Read more here.

open source

Ease of Integration

Expedite your development by reusing the many available references designs. Bindings and demo applications for C/C++, Android (Java), iOS (Swift), Python, Rust, and Web Assembly are available.

language independent

Language Independent

Can create models for almost any phrase in any language.

light bulb

This demo allows you to change the color of the smart lamp (on the left) using Picovoice's voice control engine via any of the following commands. Note that you need a working microphone. This demo is running locally in the browser. You can turn off your internet connection and it will keep working.

  • OK lamp, white.
  • OK lamp, yellow.
  • OK lamp, orange.
  • OK lamp, purple.
  • OK lamp, navy blue.

Start Demo

The speech to text SDK facilitates building conversational interfaces without the need to any cloud connection. It is powered by a novel end-to-end learning algorithm which makes it possible to perform accurate speech recognition on IoT platforms with limited memory/CPU budget.

open vocabulary large vocabulary

Open Vocabulary

No limit on the size of vocabulary. Provides large vocabulary transcription capabilities on embedded platforms.

continuous real-time

Continuous and Real-Time

Transcribes audio in real time. Reduces latency and improves the user experience.

lightweight efficient IoT embedded

Lightweight

Runs in real-time with only 5.6 MB of memory and 25% CPU usage on a Raspberry Pi 3. Natural fit for IoT platforms. Android, iOS, Raspberry Pi, and a growing number of IoT platforms are supported.

Picovoice's speech to intent SDK enables transforming natural speech into actionable structured data. It is the fusion of our speech to text engine with an efficient domain-specific natural language understanding engine. The speech to intent engine can perform natural speech transcription and understanding using limited CPU/memory budget.

continuous real-time

Continuous and Real-Time

Transcribes and understands speech in real time. Reduces latency and improves the user experience.

lightweight efficient IoT embedded

Lightweight

Runs in real-time with less than 2 MB of memory and 8% CPU usage on a Raspberry Pi 3. Natural fit for IoT platforms.

Picovoice is a team of applied scientists and engineers who strive to build a future where our lives are enhanced with ambient voice AIs, while respecting your privacy. Picovoice is founded by Alireza Kenarsari. Prior to Picovoice Alireza was a Senior Engineer at Amazon and has been also an early-stage engineer in a few successful technology startups (one reached IPO). He is the inventor of five US patents within the fields of deep learning and speech recognition. Read more about the beginnings of Picovoice here.

[August 26, 2018] Picovoice's speech to text repository ranks among top 10 open source machine learning projects. We are extremely excited to be on this list alongside names such as OpenAI, NVIDIA, Facebook Research, Airbnb, and Google. Read more here.

[July 23, 2018] Picovoice ranks among top 10 open source machine learning projects. We are extremely excited to be on this list alongside names such as Facebook Research, Salesforce, Baidu Research, and NVIDIA. Read more here.