LenguASR - Realtime*, Offline Speech Recognition for Unreal Engine

LenguASR delivers a voice interface that just works. No need for cloud services, long setup or extra plugins. Inference runs on the CPU, freeing Players' GPUs free for what they are meant for - rendering graphics.

LenguASR leverages Nemotron 3.5 ASR to power multilingual real-time transcription on standard x64 CPUs. Set up is easy: drop in the Speech Component on a Blueprint Actor, preload the model, then start capturing mic audio and handle partial or final text events.

• Integrated Microphone and VAD: No separate audio handling is required. Silero VAD automatically detects speech intervals in continuous mode. Alternatively, you can use push-to-talk for more controlled activation.
• Performant: Offers multi-core support. 2-3 cores (running background threads) are sufficient for continuous voice recognition on modern PC CPUs, as well as consoles PS5 | Xbox (including Xbox Series S).

• CPU-Efficient Real-Time: Offer your players consistent frame rates by offloading speech recognition to an efficient CPU pipeline, no GPU acceleration needed. Perfect for games where the GPU is fully busy rendering.
• 32 Languages Supported: Use a single model that can detect multiple languages, from English and Spanish to Korean, and even Arabic, etc.
• Predictable Memory Footprint: Allocate about 1.25 GiB of memory per active voice recognizer (typically just 1 is required). Supports both baseline and AVX2 CPU builds, plus straightforward model loading and unloading.
• Blueprint Events & Flexibility: In continuous mode, get partial transcripts for rapid feedback. In push-to-talk mode, you control exactly when to capture speech. Use recognized text for voice commands, NPC interactions, or dynamic UI subtitles.
• One-Time Setup, Fully Offline at Runtime: Download the Nemotron 3.5 Q8 model (~743 MB) once. No API keys required, no external web services, no extra costs! Your players can stay offline and preserve their data privacy.

Included:
• Complete C++ source and a Blueprint-accessible component.
• Windows x64 native CPU libraries (baseline and AVX2).
• Editor utility to download Nemotron 3.5 Q8 and Silero VAD.
• Documentation, cvar references, and usage examples.

Supported Languages:

Transcription-ready (19 locales) English (en-US, en-GB), Spanish (es-US, es-ES), French (fr-FR, fr-CA), Italian (it-IT), Portuguese (pt-BR, pt-PT), Dutch (nl-NL), German (de-DE), Turkish (tr-TR), Russian (ru-RU), Arabic (ar-AR), Hindi (hi-IN), Japanese (ja-JP), Korean (ko-KR), Vietnamese (vi-VN), Ukrainian (uk-UA)

Broad-coverage (13 locales) Polish (pl-PL), Swedish (sv-SE), Czech (cs-CZ), Norwegian Bokmål (nb-NO), Danish (da-DK), Bulgarian (bg-BG), Finnish (fi-FI), Croatian (hr-HR), Slovak (sk-SK), Mandarin (zh-CN), Hungarian (hu-HU), Romanian (ro-RO), Estonian (et-EE)

Dependencies:

• The plugin requires Epic Games' AudioCapture Plugin to be enabled.

*The small print: Please note Realtime transcription requires a relatively modern CPU with AVX2. LenguASR runs on hardware without AVX2, but not real-time. For example, on Pentium G6600, there are about 5 seconds lag between voice detection and the first text being recognized. You can still create NPC interactions even for old CPUs without AVX2, but you will have to create design solutions around the lag.

*Console Support: The plugin runs on consoles, but this release does not include PS5 | Xbox Series X / S libraries. We can offer support integrating, but please note you have to be an approved PlayStation Partner | Xbox Developer.

🌟🌟🌟 Introductory Price Offer! 🌟🌟🌟