An official publication of the International Speech Communication Association (ISCA)

Computer Speech & Language publishes reports of original research related to the analysis, recognition, understanding, production, synthesis, coding and mining of speech and spoken language, including spoken language dialogue systems and speech-based interaction with ‘intelligent’ artefacts, and their wide-ranging applications.

The journal provides a focus for this work, and encourages an interdisciplinary approach to speech and spoken language research and technology. Thus contributions from all of the related fields are welcomed in the form of reports of theoretical or experimental studies, reviews, and pertaining to models and their implementation, or reports of fundamental research leading to the improvement of such models.

Research Areas Include

    Algorithms and models for speech recognition and synthesis
    Natural language processing for speech understanding and generation
    Computational models of spoken language discourse and dialogue
    Speech and spoken-language inclusive multimodal approaches and systems
    Information retrieval, extraction and summarization of spoken language media
    Speaker and spoken language recognition
    Computational models of speech production and perception
    Signal processing for speech analysis, enhancement and transformation
    Evaluation of speech-based interactive systems
    Applications of speech and spoken language technologies

Note that we are no longer accepting submissions devoted to pure Natural Language Processing NLP (i.e. all new manuscripts must address some aspect of spoken language processing).

Special Issue on Evaluation of speech and speech synthesis
截稿日期: 2026-06-30

Synthetic speech has advanced to a point where its quality and diversity challenge the boundaries of existing evaluation methods. Frameworks such as MOS and MUSHRA were designed to measure transmission quality rather than to assess speech as such; they were never intended to capture the communicative or functional properties of speech when transmission is no longer the limiting factor. In contemporary systems, performance ought instead to be defined by how well the speech fulfils its intended task, role, or utility.

The Special Issue therefore asks how evaluation can be made more responsive to this new landscape: one in which human and synthetic speech can, and should, be assessed by comparable principles tied to task and situation.

Much of today’s evaluation practice still relies on comparing synthetic speech to static recordings of human voices. Such tests can be useful for measuring surface similarity, but they ignore the dynamic and situational aspects that determine whether speech actually fulfils its purpose. Human speakers continuously adapt timing, prosody, and style to the communicative setting and to the role or persona they embody. A synthetic voice should be expected to perform similarly: it should use a speaking style suited to the situation or task, be it audiobook narration, dialogue interaction, public announcement, or personalised replacement voice, and align it with the intended persona, be that a robot, a disembodied assistant, a child, or an adult. This Special Issue particularly seeks evaluations that capture such situational and functional adequacy, rather than limiting comparison to perceived “human-likeness.”

Guest editors:

Prof. Jens Edlund (Executive Guest Editor), KTH Royal Institute of Technology, Stockholm, Sweden; Email: edlund@speech.kth.se 
Dr. Sébastien Le Maguer, University of Helsinki, Helsinki, Finland; Email: sebastien.lemaguer@helsinki.fi 
Dr. Christina Tånnander, MTM, Swedish Agency for Accessible Media and KTH Royal Institute of Technology, Stockholm, Sweden; Email: christina.tannander@mtm.se 
Prof. Petra Wagner, Bielefeld University, Bielefeld, Germany; Email: petra.wagner@uni-bielefeld.de 

Special issue information:

We invite contributions that reinvent, extend or refine evaluation practice in these directions, including but not limited to studies that:

• propose concrete alternatives to established evaluation paradigms, demonstrating that more informative and diagnostically useful practices are both possible and practicable;
• investigate the generalisability of established evaluation schemes across different applications or tasks, or compare various evaluation schemes within a single application domain;
• align measurement with real-world use, broadening evaluation perspectives through situated examples from accessibility, education, healthcare, entertainment, and other fields;
• provide guidance for future research, consolidating lessons into good practices and identifying the conceptual and methodological challenges that remain; or
• transfer or adapt evaluation practices from neighbouring fields such as speech therapy, HCI, or psychology.

Manuscript submission information:

Important Dates:

    Submission Open Date: December 1, 2025
    Submission Deadline: June 30, 2026
    Editorial Acceptance Deadline: March 31, 2027

Submission Guidelines:

All manuscripts should be submitted electronically through Editorial Manager® at https://www.editorialmanager.com/ycsla/default.aspx. When submitting papers, please select the Article Type as "VSI: Speech Eval & Synthesis". 

Authors should prepare their manuscripts according to the "Guide for Authors" of the Computer Speech & Language outlined at the journal website. 

All papers will be peer-reviewed following a regular reviewing procedure. 

For any further information, the authors may contact the Guest Editors.

Keywords:

speech synthesis; evaluation