A5S Async V1

A5S Async V1 is AirCaps AI’s speaker-attributed speech-to-text model for diarization-aware ASR. It is offered as an async API (a5s-async-v1) that returns transcripts with speaker labels and timestamps, aimed at meetings and overlapping speech.

AirCaps reports #1 of 15 systems on VoiceArena Diarization-Bench at about 23.89% cpWER, leadership on five public meeting benchmarks, median ~9.8 seconds to transcribe one hour of audio, and list pricing around $0.18 per audio-hour. Research writeup: https://aircaps.ai/research/a5s-async-v1-the-worlds-most-accurate-speech-recognition-model-for-speaker-attributed-transcription ; model page: https://aircaps.ai/models/a5s-async.

Highlights:

• Official AirCaps launch with independent VoiceArena Diarization-Bench #1 claim;
• Overlap-robust speaker-attributed transcription (labels + timestamps);
• Async API SKU a5s-async-v1 for batch/meeting pipelines;
• Reported ~1h → ~9.8s median turnaround at ~$0.18/audio-hour;
• Strong focus on meeting-style and multi-speaker public benchmarks.

Why it matters

Plain ASR without reliable speaker attribution breaks meeting notes, CRM call logs, and legal/compliance archives. A5S Async V1 packages diarization + recognition as one production SKU, with a public VoiceArena ranking that buyers can cross-check rather than only vendor charts.

Speed and price claims (sub-10s median for an hour of audio at low per-hour cost) matter for backlog transcription and always-on meeting bots. Overlap robustness addresses the common failure mode where speakers talk over each other in real meetings.

Limits

Benchmark leadership is snapshot-based; re-check VoiceArena and AirCaps docs for current numbers, languages, and audio formats. Async API latency is median under their test load—not a hard SLA for every file length or accent mix. cpWER and meeting WER do not guarantee domain performance (medical, heavily accented, or noisy field audio). Confirm data-retention, region, and commercial terms on aircaps.ai before sending sensitive recordings. Do not treat marketing “world’s most accurate” wording as a universal guarantee across all ASR tasks.

1. Open https://aircaps.ai/models/a5s-async and the research post to confirm model id a5s-async-v1, price, and supported languages;
2. Obtain an API key and submit audio via the async endpoint; poll or webhook for speaker-labeled transcripts;
3. Validate diarization quality on your own meeting samples (overlap, crosstalk, far-field mics);
4. Before production, review retention/privacy policy, SLA, and whether cpWER leadership still holds on VoiceArena for your use case.