Post-doc

Postdoctoral position on conversational agents for Audio Mobility 2030

This posdotoral position on conversational agents is proposed in the framework of the Audio Mobility 2030 (AM2030) project, which started in April 2023. AM2030 aims at enabling car manufacturers to have their own in-car audio application, regardless of the operating system. They will be able to deploy a global audio experience and offer the best content and proactive services to drivers. It is positioned as a true road companion that will help consumers adopt eco-responsible behaviors: vehicle self-diagnosis and maintenance reports, advice on driving and the use of on-board equipment.

Project partners: ETX Studio (Lead), Continental Automotive FRANCE SAS, Université de Toulouse – ANITI, École Polytechnique de Paris

ANITI’s role in the project is related to working on human-computer interactions, in particular on natural language understanding.  This will include a conversational model that can exploit conversational structure as well as content provided by modern transformer based models.  The model will learn constraints on the user’s preferences, from the conversation and from his previous choices.

The conversational assistant will go considerably beyond the art of current finite state dialogue systems but offering a transparency, guarantees and explainability that large transformer models by themselves cannot.  It will interact with voice based components as well as a recommendation model for actions based on the information acquired by the conversational assistant.

Required skills

Applicants should ideally have a PhD in machine learning. Good programming skills is essential; and English communication skills are also required.

Contract : post-doc

Duration : 24 months

Salary : according to experience

Advisor : Nicholas Asher

Automatic speech recognition for an in-car voice assistant

This PostDoc position is proposed in the framework of the Audio Mobility 2030 (AM2030) project, which started in April 2023. AM2030 aims at enabling car manufacturers to have their own in-car audio application, regardless of the operating system. They will be able to deploy a global audio experience and offer the best content and proactive services to drivers. It is positioned as a true road companion that will help consumers adopt eco-responsible behaviors: vehicle self-diagnosis and maintenance reports, advice on driving and the use of on-board equipment.

Project partners: ETX Studio (Lead), Continental Automotive FRANCE SAS, ANITI, Université de Toulouse, École Polytechnique de Paris.

ANITI’s role in the project is related to working on human-computer interactions, in particular on natural language understanding. The role of the hired PostDoc researcher will be to work more specifically on automatic speech (ASR, Speech-To-Text) in a noisy environment (the interior of a car). Two lines of research are envisaged: 1) adapting state-of-the-art open-source ASR models and self-supervised speech representation models (Wav2Vec2) to the noisy context of vehicles
(presence of music/radio in the background, engine noise, wind noise, rain, etc.), 2) working on the language models that constrain end2end systems. Depending on the candidate research profile, one of these research lines will be chosen,

This research will be conducted in connection
with the two other aspects treated by ANITI: 1) the study of the conversational structures between the driver and the assistant and their semantic interpretation, 2) the detection of
emotions and states of mind based on speech and transcription cues.

The hired PostDoc will be based at the Computer Science Research Institute of Toulouse (IRIT, located in the campus of the Toulouse III Paul Sabatier University. They
will be integrated in the Samova team, composed of about twenty permanent staff, PhD students and PostDocs whose research is related to various aspects of AI applied to speech and audio processing (https://www.irit.fr/SAMOVA/site/).

Required skills

Applicants should have a PhD in machine learning, ideally in speech/natural language processing.
Good programming and English communication skills are also required.

References

• Baevski, A., Zhou, H., Mohamed, A., and Auli, M. wav2vec 2.0: A framework for self-supervised
  learning of speech representations. arXiv preprint arXiv:2006.11477, 2020
  
• Radford, A., Kim, J. W., Xu, T., Brockman, G., McLeavey, C., & Sutskever, I. (2022). Robust speech
  recognition via large-scale weak supervision. arXiv preprint arXiv:2212.04356
  
• L. Gelin, M. Daniel, J. Pinquier, T. Pellegrini, 2021. End-to-end acoustic modelling for phone
  recognition of young readers. Speech Communication, 134, pp. 71-84.

Contract : post-doc

Duration : 24 months

Salary : according to experience

Location : Computer Science Research Institute of Toulouse (IRIT), Toulouse, France

Advisor : Thomas Pellegrini

Conversational sentiment analysis for an in-car voice assistant

This PostDoc position is proposed in the framework of the Audio Mobility 2030 (AM2030) project, which started in April 2023. AM2030 aims at enabling car manufacturers to have their own in-car audio application, regardless of the operating system. They will be able to deploy a global audio experience and offer the best content and proactive services to drivers. It is positioned as a true road companion that will help consumers adopt eco-responsible behaviors: vehicle self-diagnosis and maintenance reports, advice on driving and the use of on-board equipment.

Project partners: ETX Studio (Lead), Continental Automotive FRANCE SAS, ANITI, École Polytechnique de Paris.

ANITI’s role in the project is related to working on human-computer interactions, in particular on natural language understanding. The role of the hired PostDoc researcher will be to develop a Conversational Sentiment Analysis system that will be able to detect the polarity and emotion of speakers based on an ongoing interaction. Two main tasks are planned:

1. Analysis of the discursive structure of the conversation to identify speaker’s
   needs/goals/preferences.
   
2. Exploit this structure as well as past conversations to build the subjective profile of the
   speaker via the detection of the conveyed sentiments (positive vs. negative) as well as the
   emotional states of the speaker. Potential malevolent dialogues (e.g. aggressivity) will also
   be detected.
   
3. Inject this subjective profile to the recommender system to increase its performances.
   The hired PostDoc will be based at the Computer Science Research Institute of Toulouse (IRIT,
   https://www.irit.fr/en/), located in the campus of the Toulouse III Paul Sabatier University. They will be integrated in the Melodi team, composed of about 30 permanent staff, PhD students and PostDocs whose research is related to various aspects of AI applied to text and dialogue processing (https://www.irit.fr/departement/intelligence-artificielle/equipe-melodi/).

Required skills

Applicants should have a PhD in machine learning, ideally in speech/natural language processing.
Good programming and English communication skills are also required.

References

• Patricia Chiril, Endang Wahyu Pamungkas, Farah Benamara, Véronique Moriceau, Viviana Patti: Emotionally Informed Hate Speech Detection: A Multi-target Perspective. Cogn. Comput. 14(1): 322-352 (2022)
  
• Farah Benamara, Maite Taboada, Yvette Yannick Mathieu. Evaluative Language Beyond Bags of Words: Linguistic Insights and Computational Applications. Comput. Linguistics 43(1): 201-264 (2017)
  
• Farah Benamara, Nicholas Asher, Yvette Yannick Mathieu, Vladimir Popescu, Baptiste Chardon: Evaluation in Discourse: a Corpus-Based Study. Dialogue Discourse 7(1): 1-49 (2016)

Contract : post-doc

Duration : 24 months

Salary : according to experience

Location : Computer Science Research Institute of Toulouse (IRIT), Toulouse, France

Advisor : Farah Benamara

Constraint and preference acquisition for an interactive drive assistant

A postdoc position is offered within the CORAM project Audiomobility 2030, which is centered around a plan to develop an interactive assistant for drivers that will interact via voice and also visuals to help instruct and entertain passengers and driver during a car trip.

The research will be focused on acquiring a constraint model of the driver’s preferences integrating the symbolic constraints gleaned from the conversational interactions provided by another component of Audio Mobility 2030, the conversational assistant.

Project partners: ETX Studio (Lead), Continental Automotive FRANCE SAS, ANITI, École
Polytechnique de Paris

The post doc will be based at LIRMM, Montpellier, France.1
The goal of this postdoc is to design a method to acquire a constraint model representing implicit information about the driver’s (and passengers’) preferences.

This acquisition process will be based on what we know about driver/passengers (identity,
preferences, history), as well as the context (alone/accompanied, weather, short/long trip,
work/holiday…).

The proposed method may be based on prior work on the topic of constraint acquisition [Bessiere
et al. 2017]. Modeling this information by a constraint network allows the system to validate the
1 Part of the work can be done at LAAS-CNRS, Toulouse, France, depending on the recruitee’s
wishes.

Required skills

Applicants should have a PhD in artificial intelligence, ideally in constraint programming, Boolean
Satisfiability, and/or Machine Learning. Good programming and English communication skills are
also required.

References:

• Christian Bessiere, Frédéric Koriche, Nadjib Lazaar, and Barry O’Sullivan. Constraint acquisition. Artificial Intelligence, 244:315–342, 2017. Combining Constraint Solving with Mining and Learning

Contract : post-doc

Duration : 24 months

Salary : according to experience

Location : Laboratoire d’Informatique, de Robotique et de Microélectronique de Montpellier LIRMM

Advisor : Christian Bessiere and Emmanuel Hebrard