Deploy LinTO on your company’s infrastructures
We address the need for data sovereignty for actors in industry, medicine, banking, etc. More generally, we offer a system compatible with user privacy and General Data Protection Regulation (GDPR).
The platform components are designed to be deployed on your own infrastructures, independently, according to a fully scalable client-server architecture.
Are you ready for industry 4.0?
Go beyond a simple voice remote control system
- Adaptation and centralized updating of business-specific vocabulary (entity names, terms of art, proper names, etc.)
- Real-time audio transcriptions of conversations of varying length with one or more speakers
- Interoperability designed for your information system and any API driven backends
- Centralized operation and maintenance of “fleets” of voice access points
- Use in your Business to Consumer (B2C) products
- Voice client portability
A truly complete solution, ready to be deployed and used
A full integration of leading speech recognition tools in your workflows means that LinTO’s features can be deployed effectively and efficiently.
|LinTO||GAFAM / IBM / Nuance||Snips||MyCroft|
|Open Source||Complete, free||Some parts||Some parts|
|Speech To Text||*On premise||Cloud||Embedded only||Cloud (GAFAM / IBM)|
|API access||Extra fee|
|B2C use||Extra fee||For simple remotes|
|Interoperability with your APIs and data sources||Limited|
|Exploitation, monitoring fleet||Partial|
|Multiplatform client||Raspberry Pi, Android system and Web (coming soon)||Embedded systems, Raspberry Pi||Embedded systems, Raspberry Pi|
|Professional use||Not recommended||Not recommended|
*On premise: In some cases (simple command mode), LinTO's Speech To Text service can be embedded in the target hardware device; please contact us for further information. LinTO also offers a Wake Word feature that is embedded in the voice client.
Exciting features coming soon
The product roadmap is the result of an ambitious research project and is centered around the development of large vocabulary, multimodal and multi-speaker functionalities.
Drawing on recent advances in AI and machine learning, we will soon offer new tools to facilitate the production of meeting minutes and summaries, including complete multi-speaker transcription, speaker identification, and visual (video-based) recognition of participants.
Our publications on these topics:
- Data Programming for learning discourse structure
- Conference : ACL 2019
- Apprentissage faiblement supervisé de la structure discursive
- Conference : TALN-RECITAL 2019
- Learning multi-party discourse structure using weak supervision
- Conference : Dialogue 2019
- Weak Supervision for Learning Discourse Structure
- Conference : EMNLP 2019
- Char+CV-CTC: combining graphemes and consonant/vowel units for CTC-based ASR using Multitask Learning
- Conference : Interspeech 2019
- Unsupervised Abstractive Meeting Summarization with Multi-Sentence Compression and Budgeted Submodular Maximization
- Conference : ACL 2018
- Lexical Emphasis Detection in Spoken French Using F-BANKs and Neural Networks
- Conference : SLSP 2017
LinTO in your company
Modular client-server architecture and services
The LinTO platform is made of independent services and usable as you wish in a client / server logic
LinTO Client Connection Service: LinTO - Server
This component is the secure entry point for all the client systems of your platform
It uses the standard IoT MQTT to ensure real-time operation with perfect service quality
Any software with MQTT or WebSocket connectivity can establish a connection with your platform
Speech transcription service: LinTO - LinSTT
The cornerstone of the LinTO platform, LinTO - LinSTT transcribes audio provided via an API
- Usable via streaming or file-upload
- Can be integrated into other workflows within the company, for example, to add voice access to existing business applications or chatbots
- Responds entirely to the industrial challenges of deployment and service quality with a solid scalable implementation on clusters and a management API
- Transmits results in text format for audio files of varying length
- Adaptable to specific use cases using specialized AI models
- Effectively transcribes proper names and business terms of art, and provides context-specific formatting (dates, numbers, acronyms...)
- Automatically learns from your usage when you add or remove skills from your platform
Outstanding PerformancesExpressed as error rate (Word Error Rate)
|Read speech||Spontaneous speech|
- Corpus selection:
- The results are based on two corpora, one of read speech and one of spontaneous speech, each consisting of 30 minutes of selected audio segments paired with their verbatim transcriptions.
- All recordings were carried out in a clean environment ("SNR ~= 20").
- The corpus of read speech consists of 8 speakers (4 men and 4 women) reading a French newspaper.
- The corpus of spontaneous speech comes from a radio program involving 15 male speakers.
- Calculation of performance and normalization:
- Performance was evaluated on the verbatim transcription without punctuation following a normalization of numbers, dates, percentages, etc.
- The Word Error Rate (WER) reported in the table measures the amount of insertion, suppression, or word substitution predicted relative to the normalized transcripts. The lower the value, the higher the system’s performance.
Natural Language Understanding system: LinTO - NLU
This component evaluates and understands natural language provided to its API in text format. It detects intentions, resolves entities and produces a JSON response that can then be used by the skills execution service.
- Based on the SNCF Tock project (The Open Conversation Kit)
- Supports other engines such as Rasa or Stanford NLP
- Supports the use of several AI models dedicated to specific use cases
- Automatically learns according to your usage when adding or removing skills from your platform
Skills Execution service: LinTO - Skill Server
Based on IBM's Node Red it allows the implementation of skills for one or more LinTO customers in a given business context (a meeting room, a corridor or even as part of the management of a voice application).
- Compatible with all skills in the existing Node Red skills base (more than 2000!)
- Centralized configuration of data bus systems, authentication information, etc.
- Straightforward development of custom business skills with many examples provided - see documentation
- Automatic re-training of AI models when deploying or deleting skills and dictionaries
- Easy interoperability of skills with all your APIs or connected objects through visual workflow programming
LinTO Administration Interface: LinTO - Admin
All configuration operations and platform maintenance are handled through a web application.
- Provides templates to define specific workflows for each context that you need to manage (meeting room, application, etc.)
- Offers visual workflow editing (adding/removing skills)
- Manages updates and operation of device fleets
- Permits real-time monitoring of connected LinTO devices
- Allows for provisioning and deployment of LinTO devices by serial number
- Offers remote control of volume, network connectivity, languages, etc.
Example of LinTO Deployment on premise
LinTO as part of your solution
LinTO for Raspberry Pi & ARM systems
We currently offer a standard LinTO device that runs on the Raspberry Pi (or other ARM smart boards). This option is ideal for demonstrating your R&D work on automatic speech recognition in business.
This is a turnkey solution available from our developers website in the form of a Raspberry kit, guides and download links.
For makers, hobbyists and R&D teams who want to free themselves from any dependence on GAFAM systems, we offer the most advanced and state-of-the-art system for open-source voice assistant projects.
- Customizable embedded Wake Word(s) models
- Voice processing chain contained entirely on clean infrastructures
- Complete suite of ARM system image creation tools based on the Qemu emulator
- Rich and easily adaptable graphical interface
LINAGORA offers complete project management assistance in order to help you move from a prototype to production of a fleet of devices that meet your specific criteria.
Would you like to design a smart-assistant?
In this growing market, many players such as MyCroft, Archos or Sonos, offer solutions built around Google Home or Alexa.
However, certain characteristics of these systems - no data protection and no business vocabulary adaptation - limit them to a B2C market that is not (yet) concerned by data sensitivity and criticality issues.
The LinTO platform embraces these challenges from the start in order to be the engine that catapults your professional product.
The connection of a device system to the platform is simplified by using the LinTO - Server component that supports the standard IoT protocol, MQTT. This connection is easily implemented in the development of software dedicated to your project.
In order to streamline the LinTO experience, the development of LinTO applications for Android and web are front and center on our product roadmap.
This multi-platform device is currently under development and aims to deliver against industry expectations for solution deployment without specialized development (first public version target Q2 2020).
Android voice assistant WITHOUT Google
These initiatives are based on various alternative APIs for Google products (example)
Alternatives exist for push notifications, maps, the search engine, etc., but there are none to replace the famous "Ok, Google".
Many companies distribute professional Android products through this type of AOSP (Android Open Source Project) distribution but are in a deadlock with respect to voice-operated solutions.
The LinTO platform aims to finally implement voice interactivity in an Android mobile device that fully respects user data privacy. If you see this as the right solution for your company, contact us for an overview of our system capabilities.