Anam Transcription

Hybrid realtime + canonical transcription

Disconnected OpenAPI

1 · Connect

Paste the API key printed by make seed. It is sent in the Socket.IO handshake and in the Authorization header — nothing is stored beyond this tab's session.

These start from the account when you connect or save Settings. Unchecking one applies to this recording only and is not written back.

2 · Record

Modo de captura

Contagem por faixa de áudio. Automático deixa o pyannote estimar; N fixo vira num_speakers naquela faixa.

Presencial: só o microfone. Online: escolha uma aba do Chrome com áudio da aba + mic.

Ao conectar, o navegador pode pedir acesso ao microfone para listar os dispositivos com nome — escolha o mic antes de Iniciar. Troca mid-session é atômica (ideal deviceId).

Elapsed
0.0s
Sent
0 KB
Engine
Windows
0
Job
—
Status
idle
Fase
idle

Pausar para a captura sem fechar o job (dá para Continuar depois). Gerar relatório finaliza e enfileira a transcrição canônica. Fechar a aba pausa — não finaliza.

Transcrição realtime

Fonte: realtime Aguardando gravação — partials ao vivo aparecem aqui.

Nenhuma gravação ativa. Ao gravar, o texto parcial (rascunho) aparece aqui em tempo quase real. Falantes e diarização ficam só na aba Canônico.

3 · Settings

Account defaults and caps for this API key, saved with PATCH /v1/me. Connect’s checkboxes are this recording only and cannot exceed these. A blank limit inherits the deployment default.

Where we POST job events. Signed with your webhook secret, retried with backoff. Leave empty to poll GET /v1/jobs/:id instead.

priority (premium) — finalisation goes to a dedicated queue with its own worker, so the canonical result lands seconds after a session ends. queued (basic) — the shared queue: same pipeline and same quality, but it may wait behind other tenants' work.

Features · account

Capabilities for every recording on this key. Connect does not edit these: “This recording · live partials” and “This recording · speaker diarization” only choose what the next session asks for, inside what is enabled here.

Allows live partial text. Off means no recording on this key gets a live view, and nothing is charged for one. The Connect checkbox cannot turn this on.

Also POST live partials to your webhook. Chatty: one delivery per update. Most integrations read partials over the socket and leave this off. Not the same as the Connect live-partials checkbox.

Allows “who spoke when” on the canonical result. Off means this key never diarizes, and Connect starts unchecked. On presets Connect checked; unchecking there skips diarization for one recording and is not saved here.

How many recordings this tenant may have open at once. Reaching it does not refuse a recording; it turns off the live view for the extra ones.

Limits · account caps

Caps for this tenant, not for the recording you are about to start. Each box shows the value in force. deployment default means it follows the server and will keep following it; editing pins it to this account. Clear a box to hand it back.

Longest recording this account accepts. A 50-minute session is 3000. The server default is 14400 (4 hours) and that value can be saved. Longer uploads are rejected at ingest rather than part-way through.

Per-file ceiling on REST uploads. 16 kHz mono WAV is about 115 MB per hour, so the default of 524288000 is roughly four and a half hours of it.

Rate limit on job creation and session starts. Exceeding it returns 429 with Retry-After; it bounds how fast you enqueue, not how much you may store.

How much audio each live transport frame carries. Lower means partials arrive sooner but the engine decodes more often, so it costs more GPU and falls behind more easily on a CPU. Higher is cheaper and steadier, with later text. 480 is the balance point.

How long a live session may send no audio before the server closes it and transcribes what arrived. This is what stops a dropped client from leaving a recording open forever; a client that reconnects before it elapses resumes the same recording.

Event log