Picovoice
Picovoice is a private Canadian AI company that provides 15 on-device SDKs for voice, language, and vision understanding to Fortune 500 enterprises in automotive, aerospace, healthcare, and consumer electronics, eliminating cloud dependency through proprietary training, compression, and inference technology.
- Company typePrivate
- Founded2019
- HeadquartersVancouver, Canada
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What Picovoice does
Picovoice is a privately held Canadian AI company that develops on-device artificial intelligence software development kits (SDKs) for real-time voice, language, and vision understanding. The company's technology stack is purpose-built for edge deployment, comprising a proprietary pipeline—picoGym for model training, picoCompression for model compression, and picoInference for runtime execution—with no dependencies on third-party frameworks such as PyTorch, ONNX, or TensorFlow. Picovoice offers 15 production-ready SDKs spanning voice activity detection (Cobra), wake word recognition (Porcupine), speech-to-intent (Rhino), streaming and batch speech-to-text (Cheetah, Leopard), text-to-speech (Orca), speaker recognition and diarization (Eagle, Falcon), spoken language identification (Bat), noise suppression (Koala), neural machine translation (Zebra), on-device large language model inference (picoLLM), optical character recognition (Alpaca), and vision-language models (Falcon). Each SDK runs entirely on-device, eliminating cloud API calls, network latency, and outbound data transmission. The company is headquartered in Toronto, Canada (firmographic data lists Vancouver), and serves Fortune 500 customers including NASA, Adobe, Meta, LG, Ferrari, Netflix, Motorola, Accenture, ArianeGroup, Analog Devices, HCL, Woven by Toyota, Cox Automotive, SaskTel, Baxter, and Stanford, spanning verticals such as aerospace, automotive, healthcare, telecommunications, consumer electronics, semiconductors, and consulting. The company trains approximately 250,000 custom wake words annually and markets itself as trusted by 75+ Fortune 500 companies.
Picovoice firmographics
Firmographics- Name
- Picovoice
- Legal name
- Picovoice Inc.
- Website
- https://picovoice.ai
- Company type
- Private
- Founded year
- 2019
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Picovoice is a private Canadian AI company that provides 15 on-device SDKs for voice, language, and vision understanding to Fortune 500 enterprises in automotive, aerospace, healthcare, and consumer electronics, eliminating cloud dependency through proprietary training, compression, and inference technology.
- Ownership category
- akta.pro rank
Picovoice industry classification
Industry- Product category
- On-Device Voice AI Platform
- NAICS
- Software Publishers (5132), Computer Systems Design and Related Services (5415)
- SIC
- Services-Prepackaged Software (7372), Services-Computer Programming, Data Processing, Etc. (7370)
- akta.pro primary industry
- On-Device Inference Runtimes & SDKs (mobile/embedded) (HDAAAJAB)
- akta.pro secondary industries
- On-Device Speech & Audio AI (wake word, ASR, enhancement) (HDAAAJAF), AI Application Enablement Platforms (Copilot/Agent Frameworks, SDKs) (HDAEANAJ), Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem) (HDAEANAC), Model Deployment, Serving & Inference Platforms (HDAAABAF)
Keywords
Where Picovoice is headquartered
LocationHeadquarters
- HQ city
- Vancouver
- HQ country
- Canada
- HQ region
- North America
Offices1 record
Markets served
Picovoice business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Personnel, Technology or R&D, Marketing or Sales, Infrastructure, Operations
Revenue model
- SDK Licensing (Enterprise): Enterprise customers license production-grade on-device AI SDKs with flexible deployment licensing, dedicated engineering support, and SLA-backed response times. Custom model training available through NRE engagements.
- Free Tier with Self-Serve Access: Developers can start building with free trial and self-service access to Picovoice Console for training and deploying models. No credit card required.
- Enterprise Support Contracts: Dedicated engineering support and white-glove service for production deployments. SLA-backed response times. NDA-protected custom model training.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Freemium | Monthly | Free Trial - Self-serve developer access |
| Subscription | Annual | Enterprise - Custom licensing and support |
Go-to-market motion2 records
Distribution channels3 records
Marketing channels5 records
Picovoice product offering
Product offeringCore offering
Picovoice develops and licenses on-device AI SDKs for real-time voice, language, and vision understanding, including voice activity detection, wake word spotting, speech-to-text, text-to-speech, speaker recognition, language identification, noise suppression, on-device LLM inference, OCR, and neural machine translation. Each SDK ships as a self-contained package with no cloud API calls, no network latency, and no data leaving the device, targeting embedded, mobile, desktop, server, and web browser environments. The company owns the full AI pipeline (training via picoGym, compression via picoCompression, inference via picoInference) purpose-built for edge deployment, with no third-party framework dependencies.
Product overview
Picovoice is an on-device AI platform offering 15 production-ready SDKs across three modality categories — voice (Cobra VAD, Porcupine Wake Word, Rhino Speech-to-Intent, Cheetah Streaming STT, Leopard STT, Orca Streaming TTS, Eagle Speaker Recognition, Falcon Speaker Diarization, Bat Spoken Language ID, Koala Noise Suppression), language (Zebra Translate, picoLLM LLM), and vision (Alpaca OCR, Falcon Vision-Language Models) — supported by three core technology layers (picoGym training, picoCompression, picoInference). Each SDK ships as a self-contained, self-hosted package requiring no cloud API calls, no network connectivity, and no data leaving the device. The platform targets embedded (MCU/MPU), mobile, desktop, server, and web browser environments with cross-platform SDKs in Python, Node.js, Android, iOS, React, Flutter, React Native, .NET, Java, C, and Web. The technology stack is built end-to-end by Picovoice with no third-party framework dependencies (no PyTorch, ONNX, TensorFlow), enabling unique performance benchmarks versus cloud and open-source alternatives. Picovoice's products are designed for enterprise deployment with flexible licensing, dedicated engineering support, NRE custom model training, and SLA-backed response times.
Differentiator
Problem solved
Functional benefit
Brands
- Cobra: Voice Activity Detection SDK - highly accurate and efficient voice activity detection for on-device deployment
- Porcupine
- Rhino
- Cheetah
- Leopard
- Orca
- Eagle
- Falcon
- Koala
- Bat
- Zebra
- picoLLM
- picoGym
- picoCompression
- picoInference
Products and services
- Cobra Voice Activity Detection On-device voice activity detection (VAD) engine that detects the presence of human speech in real-time audio streams with 98.9% true positive rate at 5% false positive rate at 0dB SNR. Cross-platform SDKs for Android, iOS, C, .NET, Node.js, Python, and Web with 450KB total library size and 3.7% CPU usage on Raspberry Pi Zero.
- Porcupine Wake Word On-device wake word (keyword spotting) engine that listens continuously for specific trigger phrases to activate voice-enabled applications. Achieves 97.3% accuracy at 1 false alarm per 10 hours at 10 dB SNR, with 0.6% CPU usage on Raspberry Pi 5. Supports custom wake word training via Picovoice Console in seconds with no ML expertise required. Supports up to 24 keywords across 8 languages simultaneously.
- Rhino Speech-to-Intent On-device speech-to-intent engine that infers user intent directly from spoken commands within defined domains, bypassing the need for full STT + cloud NLU pipelines. Enables context-aware voice command and control with customizable intents and slots across Android, iOS, C, .NET, Flutter, Java, Node.js, Python, React, React Native, microcontrollers, Raspberry Pi, and Web.
- Cheetah Streaming Speech-to-Text On-device real-time streaming speech-to-text engine that transcribes speech as it is spoken with partial and final transcript emission word-by-word. Achieves 10.1% WER in English versus 11.9% for Google Streaming STT, with 590ms word emission latency and 40x less compute than Moonshine. Supports English, French, German, Italian, Portuguese, and Spanish with 34MB model size.
- Leopard Speech-to-Text On-device batch speech-to-text engine for transcribing audio and video recordings to text with embedded speaker diarization, custom vocabulary, word-level timestamps, confidence scores, and automatic punctuation. Achieves 9.7% WER in English with 0.026 core-hour ratio (12x more efficient than Whisper Base). Runs on CPU-only without GPU across 8 languages.
- Orca Streaming Text-to-Speech On-device streaming text-to-speech engine that synthesizes speech from text with guaranteed response time by eliminating network latency. Reads streaming LLM responses as they emerge for fast voice interactions. Supports cross-platform deployment across Linux, macOS, Windows, Android, iOS, Node.js, Python, Web, and Raspberry Pi.
- Eagle Speaker Recognition On-device speaker recognition (voice biometrics) engine that identifies or verifies speakers by their unique vocal characteristics. Achieves 0.18% Equal Error Rate on VoxConverse (3x lower than SpeechBrain, 4x lower than pyannote). Supports speaker verification (1:1) and identification (1:N) modes and is used in personalized wake word pipelines to gate voice activation to enrolled users.
- Falcon Speaker Diarization On-device speaker diarization engine that segments audio by speaker, identifying who spoke when. The cross-platform speaker diarization solution works with any speech-to-text engine and is embedded within Leopard Speech-to-Text or available as a standalone SDK.
- Bat Spoken Language Identification On-device spoken language identification engine that identifies the language being spoken in an audio stream in real time. Achieves 92.86% accuracy with 62x less memory and 9x less CPU than alternatives, enabling multilingual on-device voice applications and automatic language switching.
- Koala Noise Suppression On-device noise suppression engine that removes noise from speech audio in real time, enhancing audio quality for downstream voice AI processing. Available across Android, iOS, C, Linux, macOS, Python, Raspberry Pi, Web, and Windows.
- Zebra Translation On-device neural machine translation engine enabling real-time translation without cloud dependency. Provides cloud-level translation accuracy at 2.4x faster speed while using 1/6th the memory compared to Helsinki-NLP/opus-mt. Supports Android, iOS, C, Linux, macOS, Python, Raspberry Pi, Web, and Windows.
- picoLLM On-device large language model inference engine that runs X-bit quantized LLMs entirely on-device across Linux, macOS, Windows, Android, iOS, Chrome, Safari, Edge, Firefox, Raspberry Pi, and embedded platforms, supporting both CPU and GPU. Powers conversational AI, document QA with RAG, call assist, and voice memo applications with no cloud dependency.
- Alpaca OCR On-device optical character recognition engine that extracts text from documents and images, part of Picovoice's vision AI product line. Processes image input from documents and visual media to extract text entirely on-device.
- Falcon Vision-Language Models On-device vision-language models that process and understand visual inputs combined with language understanding, completing Picovoice's vision AI product line alongside Alpaca OCR. Enables image understanding combined with language capabilities for end-to-end on-device AI applications.
Quantifiable outcome
- 98.9% VAD accuracy at 5% false positive rate (12x better than Silero at 87.7%)
- +5 more outcomes
Companies that use Picovoice
Customer profileNamed customers15 records
Segments4 records
Ideal customer profiles4 records
Picovoice technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability8 records
Feature11 records
Picovoice partnerships and signals
Strategic signalScale indicators3 records
Recent moves6 records
Expansion highlights6 records
Picovoice competitors and assessment
Company assessmentDirect peers
- Sensory: Sensory is a long-standing provider of on-device speech recognition, wake word, voice biometrics, and sound identification SDKs for embedded and consumer hardware — the closest direct competitor to Picovoice's voice-first edge AI portfolio across IoT, automotive, and consumer electronics.
- Cerence AI: Cerence builds AI-powered in-vehicle voice, gesture, and driver monitoring assistants for the automotive industry — a direct competitor in on-device voice AI for connected cars, where Picovoice lists Ferrari, Woven by Toyota, and Cox Automotive as customers.
- SoundHound: SoundHound offers Houndify, an independent voice AI platform with speech-to-text, natural language understanding, and wake word across automotive, restaurant drive-thru, and IoT — overlapping directly with Picovoice's enterprise voice SDK portfolio and target verticals.
- Deepgram: Deepgram provides enterprise-grade speech-to-text, voice agent, and speech analytics APIs with claimed accuracy and latency leadership — a direct competitor to Picovoice's Cheetah/Leopard STT products for enterprise transcription and voice agent workloads.
- AssemblyAI: AssemblyAI offers speech-to-text, speech understanding (diarization, sentiment, summarization) and LLM-powered audio intelligence via API — a direct competitor in the enterprise speech AI market, particularly for batch transcription use cases competing with Leopard.
- Speechmatics: Speechmatics provides automatic speech recognition with broad language coverage and on-premise deployment options for enterprise — a direct competitor in privacy-sensitive enterprise STT where Picovoice's Leopard competes on cost and language breadth.
- Syntiant: Syntiant produces ultra-low-power neural decision processors with on-device audio ML for wake word, voice control, and sensor fusion — a direct competitor in microcontroller-class always-listening voice AI where Picovoice's Porcupine and Cobra compete for embedded designs.
Broad incumbents
- NVIDIA Riva: NVIDIA Riva is a GPU-accelerated speech and translation AI SDK for on-prem and edge deployments, leveraging NVIDIA's hardware stack — a broader incumbent offering overlapping capability with Picovoice's STT/TTS/translation on edge devices powered by NVIDIA silicon.
- Apple (Siri / Private Cloud Compute): Apple ships on-device wake word, STT, and increasingly on-device LLM capabilities across iOS/macOS via Siri and Apple Intelligence — a broad incumbent setting the user experience baseline for on-device voice AI that Picovoice's SDK partners must match or exceed.
Emerging players
- OpenAI Whisper: Whisper is the open-weight speech recognition model explicitly cited in Picovoice's Leopard benchmarks — an emerging open-source competitor whose accuracy improvements and community derivatives (Whisper.cpp, Distil-Whisper) define the open-source bar that Picovoice must beat on cost/efficiency.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
Picovoice social profiles
Digital presencePicovoice compliance and trust
Trust signalCompliance5 records
Picovoice financial estimates
Financial estimateRevenue estimate
Valuation estimate
Picovoice leadership team
Management profileNumber of profiles
Profiles1 record
Picovoice funding detail
Funding detailFunding overview
Funding rounds1 record
Investors1 record
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Picovoice M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Picovoice
What does Picovoice do?
Picovoice develops and licenses on-device AI SDKs for real-time voice, language, and vision understanding, including voice activity detection, wake word spotting, speech-to-text, text-to-speech, speaker recognition, language identification, noise suppression, on-device LLM inference, OCR, and neural machine translation. Each SDK ships as a self-contained package with no cloud API calls, no network latency, and no data leaving the device, targeting embedded, mobile, desktop, server, and web browser environments. The company owns the full AI pipeline (training via picoGym, compression via picoCompression, inference via picoInference) purpose-built for edge deployment, with no third-party framework dependencies.
Is Picovoice a public or private company?
Picovoice is a private company. It is classified as venture growth investor backed and is currently operating.
When was Picovoice founded?
Picovoice was founded in 2019. It employs 11 to 50 people.
Where is Picovoice based?
Picovoice is headquartered in Vancouver, Canada, in the North America region.
How does Picovoice make money?
Three revenue lines are on record. SDK Licensing (Enterprise) is the primary driver. The others are free Tier with Self-Serve Access and enterprise Support Contracts.
Who are Picovoice's main competitors?
Direct peers on record are Sensory, Cerence AI, SoundHound, Deepgram, AssemblyAI, Speechmatics and Syntiant. Broad incumbents are NVIDIA Riva and Apple (Siri / Private Cloud Compute). OpenAI Whisper is listed as an emerging player.
Does Picovoice have an API?
Yes. Picovoice offers multiple SDKs and APIs for on-device voice, language, and vision AI. The SDKs span Python, Node.js, Android, iOS, React, Flutter, React Native, .NET, Java, C, and Web across voice (STT, TTS, VAD, wake word, speaker recognition, diarization, language ID, noise suppression), language (translation, LLM), and vision (OCR, VLM) categories. Custom vocabulary APIs allow programmatic addition of domain-specific terms and keyword boosting via cloud API. Custom model training is available via NRE engagements. Free trial available with no credit card required. Developer documentation is at picovoice.ai/docs.
What industry is Picovoice in?
Picovoice's product category is On-Device Voice AI Platform. Its primary akta.pro industry code is HDAAAJAB, On-Device Inference Runtimes & SDKs (mobile/embedded), with a secondary code of HDAAAJAF, On-Device Speech & Audio AI (wake word, ASR, enhancement). Its NAICS code is 5132 and its SIC code is 7372.