POST /speak

Legacy, unversioned demo route. New integrations should use Transcriptions, Speech, and Translate instead.

{
  "audio_b64": "<base64 audio>",
  "direction": "grt-text" | "grt-en"
}

`direction`	Result
`grt-text`	Garo transcript only (`src_text`)
`grt-en`	Garo transcript + English text + English WAV (`audio_b64`)

Demo builds may write input audio and transcripts under a log directory. Licensed production images default logging off.