capability: speech.transcription
version: 1
description: Offline bounded speech-to-text transcription.
input_schema:
  type: object
  additionalProperties: false
  required: [audio_base64, media_type]
  properties:
    audio_base64: {type: string, minLength: 4, maxLength: 22369624}
    media_type: {const: audio/wav}
    language: {enum: [nl, en, fr, auto]}
output_schema:
  type: object
  additionalProperties: false
  required: [text, language, duration_seconds]
  properties:
    text: {type: string}
    language: {type: string}
    duration_seconds: {type: number}
    segments: {type: array}
modalities: {input: [audio], output: [text, json]}
languages: [nl, en, fr]
streaming: false
structured_output: true
quality_metrics: [wer, real_time_factor]
upgrade_class: behavioral
fallback: {allowed: false, mode: hard_fail}
privacy: {classification: restricted, allow_persistence: false, allow_logging_payloads: false, allow_network_egress: false}
resources: {accelerator_required: true, cpu_fallback_allowed: false}
production_priority: background
default_residency: load_on_demand
estate:
  category: AUDIO
  purpose: Private local transcription of bounded WAV audio without cloud fallback.
  stability: experimental
  resource_class: HEAVY
  evaluation_type: asr
  consumers: []
  payload_limits: {max_bytes: 16777216, max_batch_count: 1, max_duration_seconds: 120}