Large-Vocabulary Speech Recognition
(Customizable domain full transcription ~ 200,000 word lexicon)
Embedded Large-Vocabulary Speech Recognition
RecoMadeEasy® Embedded Large-Vocabulary Speech Recognition is an embedded natural language speech recognition engine that offers comprehensive conversational voice interaction. The engine has a small memory footprint and is designed to run natively on devices that seek unconstrained natural language interfaces with high recognition accuracy in the presence of service interruption or when full, uninterrupted and secure access to a cloud server is not guaranteed.
The RecoMadeEasy® Speech Recognition engine has been developed in our research labs in New York. When presented with an audio or audio-video stream, the engine via the API returns JSON or XML results containinng the full transcript of the stream with a configurable number of contending results. The results include a likelihood score as well as a confidence score for each result. The engine also returns the timestamps of turns of audio (sentences). It is also capable of returning the timestamps for the words in the transcript, allowing for alignment and manipulation of the results alongside the original audio stream.
The engine is built to allow users to speak naturally and be
understood – even in a far-field, noisy
environment. RecoMadeEasy® is available as an SDK with an
included API that contains all necessary components for full
integration and enables engineers to get started easily and
without any work or costs for development.
The RecoMadeEasy® AudioVisual Speech Recognition engine is also available as a server-side and a standalone product.
This engine provides one of the
most accurate transcriptions for English, handling many
different dialects and accents in a single large-vocabulary
transcription engine, It is also capable of providing real-time
processing in a small memory footprint.
The speech recognition uses a streaming interface where the
recognizer, in the form of listeners and the client, both run on
the embedded device. Any light generic client capable of using a
websocket interface may stream audio/video to a listener and get
back real-time results of the transcript with optional
alternative results, including likelihood scores in any codec
that is supported by GStreamer-1.0, including MP3, Ogg Vorbis,
Free Lossless Audio Codec (FLAC), MP4, Pulse Code Modulation
(PCM), or other codecs such as those supported by a standard
Waveform Audio File Format (WAVE).
Supported Operating Systems
The RecoMadeEasy® Embedded Speech Recognition engine
is available for the following operating
systems. The C++ SDK, command-line interface, and web
services may be used in any of the following systems:
Embedded Operating Systems:
- Fedora 25 Linux arm7vl and soon: Fedora 26 aarch64 (Latest)
- Ubuntu 16.04 Mate Linux (aarch64)(Latest)