Picovoice Wordmark
Picovoice Wordmark
Products
Technologies
Log inStart Free
Rhino Speech-to-Intent

Voice commands that work in noise and never hallucinate

Noise-robust, deterministic speech-to-intent that fuses ASR and NLU into a single on-device model. Outperforms Google Dialogflow and Amazon Lex. Private-by-design. No cloud. No latency.

Start Free
Click to activate
small
medium
large
single shot
double shot
triple shot
americano
cappuccino
espresso
6x
Higher accuracy than Big Tech average
97.3%
Accuracy tested across 6 to 24 dB Signal-to-Noise Ratio
∞
Unlimited voice interactions per user
Loved by developers, trusted by enterprises
What is Rhino Speech-to-Intent?

Deterministic voice commands for noisy, challenging environments

Most voice command systems run a two-step pipeline: speech-to-text converts audio to a transcript, then a separate NLU model parses that transcript for intent. Every step accumulates error and compounds latency. This puts user experience at risk if the product is intended to work in a factory, on a vehicle, in a hospital, or in a place without reliable connectivity.

Rhino Speech-to-Intent is different. It's an end-to-end speech-to-intent engine with a single model that maps spoken audio directly to a structured intent with typed slot values. If the spoken command doesn't fit the defined context, Rhino returns nothing. No hallucinations. No intermediate transcript. No pipeline.

Product teams define the domain, the commands that Rhino is expected to recognize, and Rhino handles the rest. It recognizes predetermined commands with high accuracy, even in noisy environments with unstable connectivity.

Developer Experience

Intent from speech in under 10 lines

A single SDK handles audio processing, model inference, and slot extraction. Rhino Speech-to-Intent provides SDKs for Python, NodeJS, Android, iOS, React, Flutter, React Native, .NET, Java, and C, enabling custom voice commands across embedded, mobile, web, desktop, and server.

OPEN-SOURCE NLU BENCHMARK
Proven accuracy in noise and across accents vs. Google Dialogflow and Amazon Lex
Rhino Speech-to-Intent is benchmarked against Google Dialogflow and Amazon Lex using 600+ real spoken commands from 50+ speakers with different accents, tested across noise levels from 6 to 24 dB SNR.
Voice Command Acceptance Accuracy
higher is better
Rhino97.3%
Amazon Lex84.3%
Google Dialogflow77.3%
* Average of 7 benchmarks at 6, 9, 12, 15, 16, 21, and 24 dB SNR Noise Levels
Voice Command Acceptance Accuracy at 21 dB SNR
higher is better
Rhino99%
Amazon Lex87%
Google Dialogflow83%
* Measured at 21 dB SNR (Low Background Noise)
Voice Command Acceptance Accuracy at 12 dB SNR
higher is better
Rhino98%
Amazon Lex85%
Google Dialogflow77%
* Measured at 12 dB SNR
Voice Command Acceptance Accuracy at 6 dB SNR
higher is better
Rhino94%
Amazon Lex76%
Google Dialogflow67%
* Measured at 6 dB SNR (High Background Noise)
See Open-source NLU Benchmark →
Ready to integrate? Check our docs to start building or talk to the sales team about enterprise deployment.
Capabilities
Why enterprises choose Rhino Speech-to-Intent
Rhino is an enterprise-ready on-device speech-to-intent engine built for applications that require accurate, noise-robust, and deterministic voice command recognition.
01End-to-end speech-to-intentTraditional pipelines run speech-to-text, then NLU — two models, two failure modes, compounding error, and latency. Rhino Speech-to-Intent is a single end-to-end model trained jointly. The unified architecture, one inference call, no transcript required, makes Rhino more accurate than chained cloud services, such as Amazon Lex and Google Dialogflow.
02Deterministic — no hallucinationsRhino Speech-to-Intent only returns an intent if the spoken command fits within the defined context. If Rhino doesn't understand, it returns isUnderstood: null, not a hallucinated guess. For safety-critical applications like medical devices, industrial equipment, and automotive interfaces, this determinism is non-negotiable.
Click to activate
small
medium
large
single shot
double shot
triple shot
americano
cappuccino
espresso
03Custom voice commands with no ML expertiseSelf-service Picovoice Console allows enterprises to define their domain using a simple YAML grammar and train custom models at any time, enabling fast iterations. Training models with specific intents, slot types, and expressions is as simple as just typing.
04Dynamic Slots with Rhino Speech-to-Intent APIThe Rhino Speech-to-Intent API lets developers and end users define custom slots and train personalized models from any device via cloud API, with models optimized for the target platform at download time.
05Dedicated Model TrainingPicovoice researchers work on Non-Recurring Engineering (NRE) engagements and fine-tune custom AI models for any target hardware platform, acoustic environment, speaker accent, and noise characteristics.
06Lightweight — Under 2.5 MBThe total storage required for Rhino Speech-to-Intent inference runtime and context files is less than 2.5 MB. With minimal compute requirements, Rhino enables complex voice commands in real-time applications even on compute-constrained devices, while leaving CPU and memory for the other components.
07Cross-Platform Speech-to-IntentRhino Speech-to-Intent runs on every platform your product ships — Android, Arm Cortex-M, Chrome, Edge, Firefox, iOS, Linux, macOS, Raspberry Pi, Safari, and Windows — across AMD, Intel, NVIDIA, and Qualcomm hardware.
08Private by DesignRhino Speech-to-Intent processes audio on the device, meaning that audio never leaves the device. No microphone data is transmitted, no cloud logs are created, and no third-party data retention occurs. GDPR, HIPAA, and CCPA compliant by design — not by policy.
09Tunable endpointingRhino Speech-to-Intent's tunable endpointing prevents cutting off users mid-sentence while avoiding awkward delays. This flexibility allows enterprises to adjust their applications based on user behavior. Wait for slow speakers and capture commands fully: "Set the temperature to 25 Celsius [short pause] in the living room", or act fast: "Move the robot arm by 5 degrees."
10Enterprise ReadyRhino Speech-to-Intent is production-grade and enterprise-ready. Picovoice offers flexible licensing, dedicated engineering support, NDA-protected custom model training, and SLA-backed response times for teams shipping at scale.
Ship it.
On device.
Reliable, accurate, and lightweight voice command detection
Start Building
FAQ
Common questions about speech-to-intent
+
What are the use cases and applications of Natural Language Understanding?
Enterprise Voice Automation
  • Manufacturing floor voice commands
  • Warehouse management systems
  • Quality control voice interfaces
  • Industrial equipment control
Healthcare Applications
  • Medical device voice control
  • Patient data entry systems
  • Hands-free clinical workflows
  • HIPAA-compliant voice interfaces
Smart Home and IoT
  • Voice-controlled appliances
  • Home automation systems
  • IoT device integration
  • Custom voice assistants and AI agents
Automotive and Transportation
  • In-vehicle voice commands
  • Fleet management systems
  • Navigation voice control
  • Driver assistance interfaces
+
What is Natural Language Understanding (NLU)?
Natural language understanding deals with meaning, i.e., comprehending users' intent. Researchers initially started with understanding user intents from the text. While spoken language understanding is a more specific term to refer to understanding user intent from speech, many people, including the industry and researchers, still use natural language understanding for both text and speech data. This is mainly due to the conventional approach of running speech-to-text and natural language understanding engines subsequently.
+
What is intent detection?
Intent Detection is a subtask of natural language processing and a critical component of any task-oriented system. Natural language understanding solutions match users' utterances with one of the predefined classes by understanding the user's goal (i.e., intention). After matching utterances with intents, the software can initiate a task to achieve users' goals. For example, users with the intention to turn the lights off may say: "Turn the lights off.", "Switch off the lights.", "Can you please turn the lights off?". Intent detection captures the users' goal: "change the state of the lights from on to off" despite the different ways to communicate it.
+
How does speech-to-intent differ from speech-to-text?
Speech-to-text converts spoken audio into a text transcript. Speech-to-intent maps a spoken command directly to a structured intent with typed slot values — no transcript needed. Rhino Speech-to-Intent's end-to-end architecture skips the ASR-then-NLU pipeline entirely, which eliminates error accumulation between steps and significantly improves accuracy in noisy conditions. Learn more about different approaches in Spoken Language Understanding, or why Rhino is a better alternative to speech-to-text while building voice assistants.
+
Can I use Rhino Speech-to-Intent to overcome the limitations of Amazon Lex and Google Dialogflow?
Rhino Speech-to-Intent is a more accurate, resource-efficient, and faster alternative to Amazon Lex, Google DialogFlow, or other NLU engines for use-case-specific intent detection. Picovoice's Free Trial allows enterprises to evaluate Rhino Speech-to-Intent and compare it with the alternatives. However, if you're still not sure how to overcome the limitations of Amazon Lex, Google DialogFlow, and other NLU engines with Rhino Speech-to-Intent or need help with migration, Contact sales!
+
How does Rhino Speech-to-Intent differ from Natural Language Understanding (NLU) solutions such as Amazon Lex, Google DialogFlow, IBM Watson Natural Language Understanding, or Microsoft LUIS?
Rhino Speech-to-Intent -as the name suggests, converts speech into intent directly without relying on text, eliminating the need for text representation. Rhino Speech-to-Intent uses the modern end-to-end approach to infer intents and intent details directly from spoken commands. This enables developers to train jointly optimized automatic speech recognition (ASR) and natural language understanding (NLU) engines tailored to their specific domain, achieving higher accuracy.
Rhino Speech-to-Intent excels in use-case-specific applications, such as voice-enabled coffee machines or surgical robots, which involve a limited number of commands, offering high accuracy with minimal resources. In contrast, open-domain applications like voice-enabled ChatGPT handle a wide range of topics and variations. Thus, we recommend Cheetah Streaming Speech-to-Text and picoLLM for such applications.
+
How do I learn more about the terminology used for Natural Language Understanding (NLU) Engines?
Intents, expressions, and slots are commonly used in conversational AI and across various engines such as Amazon Lex, IBM Watson, Google Dialogflow, or Rasa NLU. They're used to build voice assistants or bots. You can check out the Rhino Speech-to-Intent Syntax Cheat Sheet to start building or the Picovoice Glossary to learn the terminology.
+
Does Rhino Speech-to-Intent process voice data locally on the device?
Rhino Speech-to-Intent processes voice data locally on the device.
+
Which platforms does Rhino Speech-to-Intent support?
  1. Web Browsers: Chrome, Safari, Edge, and Firefox
  2. Microcontrollers: Arm Cortex-M, STM32, and Arduino
  3. Mobile Devices: Android and iOS
  4. Desktop and Servers: Linux, macOS, and Windows
  5. Single Board Computers: Raspberry Pi
+
How do I get technical support for Rhino Speech-to-Intent?
Picovoice docs, blog, Medium posts, and GitHub are great resources to learn about voice AI, Picovoice technology, and how to start building with voice commands. Enterprise customers get dedicated support specific to their applications from Picovoice Product & Engineering teams. Reach out to your Picovoice contact or  talk to sales to discuss support options.
+
Which languages does Rhino Speech-to-Intent support?
Rhino Speech-to-Intent supports English, French, German, Italian, Japanese, Korean, Chinese (Mandarin), Portuguese, and Spanish.
+
What should I do if I need support for other languages?
Contact sales team to get a custom language model trained for your use case.
+
How can I get informed about updates and upgrades?
Version changes appear in the  and LinkedIn. Subscribing to GitHub is the best way to get notified of patch releases. If you enjoy building with Rhino Speech-to-Intent, show it by giving a GitHub star!