How to teach an AI to help assist with integrity to life-changing conversations?
Learn more about the challenge
Access the data
Q&A Session for hackers
Problem The Swiss Federal Office for Migration (SEM) on a daily basis conducts interviews with asylum applicants in order to determine their refugee status.
What’s being discussed in these interviews contains sensitive personal information – effecting in content classification as confidential. The legal issues involved are complex and in order to determine an individual's refugee status, the SEM case-workers must take into account not only the relevant internal policy guidelines, but also nuanced cultural contexts, general aspects of human communication, and credibility assessments in particular. Basis for such an assessment is an accurate transcription of everything that is being said, as well as how it is being said.
By contrast, today's widely available and popular speech to text transcription tools "smoothen" the transcribed content for easier consumption, thereby sacrificed the individual speech pattern and diction.
Objective In order to be useful for SEM, the interview transcription output must reliably reflect the recording as accurately as possible: The tool has to transcribe exactly what is being said, including mistakes such as mispronunciations, false starts, auto-corrections, grammatical errors and hesitation markers. Fillers, halting speech patterns and pauses must be visualized.
The solution handles Switzerland's official languages (German, French, and Italian), optionally also foreign languages (e.g.: Farsi, Turkish, various Arabic dialects); it correctly transcribes names of (personal, organisational and geographical), and abbreviations; highlights expressions with several different meanings; knows punctuation rules; it distinguishes between different speakers, recognizes human voices from background noise, and decreases the time required for post-editing as compared to the existing manual process.
Detailed evaluation criteria and benchmarks will be provided to the hack team.
Support for Hackers
- Data: Synthetic interview recordings in multiple (official and foreign) languages, detailed evaluation criteria based on quality and efficiency.
- People: SEM subject-matter experts available remotely (via phone, or video call) to clarify uncertainties, to discuss and evaluate transcription outputs.
Technical Preferences
- Solution must be able to handle stereo (non-channel separated speaker) voice data. The resulting transcripts should contain time markers to facilitate comparison with the interview recordings.
Why hack? Accurate asylum interview transcriptions are critical for the SEM to make fair, evidence-based decisions that uphold Switzerland's commitment to providing protection to those who truly need it while maintaining public trust in the asylum system. Enhanced transcription technology enables faster, more reliable case processing, reducing administrative burden and costs for Swiss taxpayers while ensuring consistent quality even during crisis periods when application volumes surge. Most importantly, precise documentation of refugee testimonies protects vulnerable individuals by ensuring their stories are captured faithfully, giving them the best chance for a fair assessment of their protection needs.
About the Challenge Partner
The State Secretariat for Migration (SEM) is responsible for all matters covered by legislation on foreign nationals and asylum seekers in Switzerland. Read more about our tasks.