ALBAYZIN EVALUATION CHALLENGE 2026

The “ALBAYZIN 2026 Speech Technologies for COSER: A Spanish Rural Corpus (Speech-COSER)” challenge is supported by the Spanish Thematic Network on Speech Technology (RTTH) and is organized by Universidad San Pablo CEU and AUDIAS from Universidad Autónoma de Madrid. The challenge integrates in the IberSPEECH 2026 conference that will be held in Madrid, Spain from 18th to 20th November 2026.

Challenge and database description

The challenge addresses several speech technologies on the Spanish speech rural corpus called COSER. This Spanish dataset stands out due to its substantial geographical and dialectal diversity, covering all regions in Spain, and whose speakers belong to an old population in rural areas with low education levels. This is the largest corpus of dialectal European Spanish, with recordings of rural context and senior Spanish speakers. The complexity of this dataset presents a significant challenge for speech technologies. The challenge integrates three different tracks:

  • Automatic Speech Recognition (ASR): This track focuses on automatically transcribing the audio files, for which the sequence of words that appears in each audio file must be provided.
  • Speaker Diarization (SD): This track focuses on automatically segmenting the audio files according to speaker turns, for which timestamps and speaker assignment to segments must be provided. Even if the identity of the speakers and number of speakers is unknown, the output should group segments of the same speaker under the same speaker label.
  • Spoken Term Detection (STD): This track aims to detecting a list of terms within the audio files. This list is assumed to be unknown when processing the audio. A set of occurrences for each term detected in the audio files must be generated, along with their timestamps and scores as output.

The evaluation will be performed in a continuous evaluation method, with a publicly available leaderboard where participants will be able to submit their scores for each of the tracks and check their performance with respect to other participants. This leaderboard will be frozen by the deadline evaluation submission and will be reopened for post-evaluation period.

GOLD SPONSOR

Challenge participation channels

Participants can register in any of the tracks (i.e., it is not mandatory to participate in all the tracks of the challenge) or all the tracks, depending on their choice. There are two ways to participate in the challenge according to the submission type:

  • The first way relies on editing the system description paper following the IberSPEECH 2026 paper submission template so that the submitted paper (describing the system/s and the results) will appear in the IberSPEECH 2026 proceedings following the regular peer review process (deadline set at 22nd June). Moreover, participants may also have the chance to submit an extended version of this paper to a journal. This submission way implies sending one or more representatives to the evaluation workshop, to be held in Madrid, Spain as part of IberSPEECH 2026 (November 2026), and present there upon acceptance of the paper.
  • The second way demands a free-format document in which participants describe the submitted system/s along with the results, but this will not appear in the IberSPEECH 2026 proceedings. In this case, participants are allowed to present on-line their system/s without physically attending the conference, or send a video to the evaluation organizers explaining their submitted system/s, which will be shown during the evaluation workshop. The paper submission deadline in the second submission way is September 30th, 2026 (23:59 GMT+1).

Calendar for the CHALLENGE

  • April 20th, 2026: Registration opens.
  • May 1st, 2026: Release of the training/development data.
  • June 5th, 2026: Release of the test data. System submission (leaderboard) opens.
  • July 31st, 2026: Registration deadline.
  • September 30th, 2026 (23:59, GMT +1): System submission and system description paper deadline (for system description paper according to the second system description submission way). The leaderboard will be frozen by this date.
  • October 31st, 2026: Results are distributed to the participants. The leaderboard will be reopened by this date for post-evaluation period.
  • November 18th-20th, 2026: IberSPEECH 2026 Albayzin Evaluations special session in Madrid.

More details in the evaluation plan that can be found here.

For more information, please contact Javier Tejedor and/or Alicia Lozano.

ALBAYZIN 2026 Evaluations Organizing Committee

Eduardo Lleida Solano, lleida@unizar.es, Universidad de Zaragoza, Spain

Javier Tejedor Noguerales,  javier.tejedornoguerales@ceu.es,  Universidad San Pablo CEU, Spain

Luis Javier Rodríguez Fuentes, luisjavier.rodriguez@ehu.es, Universidad del País Vasco, Spain

María del Carmen Magariños Iglesias, mariadelcarmen.magarinos@usc.es, Universidad de Santiago de Compostela, Spain