Recording and transcript storage

When an app has Recording on and its team has an S3 recording bucket configured, Delphi stores call audio and related speech artefacts for playback and investigation.

The three JSON documents

Each processed recording uses these three JSON documents:

recordings/{teamId}/{appId}/{timestamp}_{conversationId}[_{tobiConversationId}]/
├── call-transcript.json
├── transcript-stt-input.json
└── transcript-tts-output.json

Document	Contents
call-transcript.json	Call-level identity, timing, recording metadata, source stages, and the conversation snapshot.
transcript-stt-input.json	Caller turns in chronological order, with references to matching STT input audio.
transcript-tts-output.json	Bot turns in chronological order, with references to matching TTS output audio.

transcript-full.json is no longer produced because it duplicated the combined contents of the STT and TTS turn documents. Recordings created by older Delphi versions can still contain that legacy fourth file.

All three documents are UTF-8 JSON with content type application/json, ISO 8601 UTC timestamps, and a top-level schemaVersion. Consumers should ignore unknown fields so compatible fields can be added without changing the schema version.

Complete recording layout

The final output for a successfully processed call can contain:

recordings/{teamId}/{appId}/{timestamp}_{conversationId}[_{tobiConversationId}]/
├── call-transcript.json
├── transcript-stt-input.json
├── transcript-tts-output.json
├── recording-processed.mp3                 # optional processed call mix
├── recording-raw.mp3                       # optional pre-processed call mix
├── stt-input/                              # optional caller-turn audio
│   ├── 0001_client.raw
│   └── ...
└── tts-output/                             # optional bot-turn audio
    ├── 0001_bot.raw
    └── ...

The STT and TTS sequence numbers are independent, zero-padded, and start at 0001. Follow the audio or s3Audio references in the JSON instead of inferring which files exist from the folder listing. Audio and turn artefacts are optional because their presence depends on the call, flow, recording configuration, and storage-protection state.

Audio formats

• recording-processed.mp3 and recording-raw.mp3 are call-level MP3 outputs.
• Turn-level .raw files are headerless audio bytes, not WAV files. Decode them using the matching item's mediaFormat, for example raw/lpcm16_16, raw/alaw_8, or raw/mulaw_8.
• A configured per-turn MP3 output is referenced with an mp3 key. WAV is not a persisted recording-output format.

call-transcript.json

TelPhi writes this call-level document before asynchronous audio processing finishes. Its recording.recordingStatus can therefore be PROCESSING, and recording.recordingFile can still be null at upload time.

{
  "schemaVersion": 1,
  "generatedAt": "2026-07-15T10:15:30.000Z",
  "storageProtection": {
    "sensitiveInfoOnStorageObserved": false
  },
  "call": {
    "conversationId": "conversation-123",
    "callId": "call-123",
    "tobiConversationId": "tobi-123",
    "traceId": "trace-123",
    "teamId": "team-123",
    "appId": "app-123",
    "callerNumber": "+49123456789",
    "endReason": "hangup",
    "startTime": "2026-07-15T10:15:00.000Z",
    "endTime": "2026-07-15T10:15:30.000Z",
    "durationSeconds": 30,
    "flowDefinitionId": "flow-123",
    "flowVersion": 4
  },
  "recording": {
    "bucket": "recording-bucket",
    "region": "eu-central-1",
    "s3Prefix": "recordings/team-123/app-123/example",
    "transcriptS3Key": "recordings/team-123/app-123/example/call-transcript.json",
    "recordingStatus": "PROCESSING",
    "recordingFile": null,
    "stages": []
  },
  "transcript": [],
  "nodeHistory": [],
  "actionsLog": [],
  "finalContext": {},
  "systemContext": {}
}

The transcript, nodeHistory, actionsLog, finalContext, and systemContext values are conversation snapshots. The recording.stages array describes the available source tracks and their captured turn segments. A stage is pre-audioproc, post-audioproc, or outbound.

STT and TTS turn documents

transcript-stt-input.json contains only caller (client) turns. transcript-tts-output.json contains only bot turns. They share this envelope and item format:

{
  "schemaVersion": 1,
  "generatedAt": "2026-07-15T10:15:31.000Z",
  "call": {
    "conversationId": "conversation-123",
    "callId": "call-123",
    "tobiConversationId": "tobi-123"
  },
  "items": [
    {
      "turnId": "stt-1",
      "side": "Client",
      "speaker": "client",
      "timestamp": "2026-07-15T10:15:01.200Z",
      "startMs": 1200,
      "endMs": 2200,
      "durationMs": 1000,
      "text": "Hello",
      "audio": {
        "raw": "stt-input/0001_client.raw"
      },
      "s3Audio": {
        "raw": "recordings/team-123/app-123/example/stt-input/0001_client.raw"
      },
      "mediaFormat": "raw/lpcm16_16",
      "provider": "azure_stt",
      "model": null,
      "stage": "post-audioproc",
      "nodeId": "voice-agent"
    }
  ]
}

Turn item fields

Field	Meaning
turnId	Provider or pipeline turn identifier.
side	Display label: Client or Bot.
speaker	Stable value: client or bot.
timestamp	Absolute turn start time, when available.
startMs	Turn start relative to recorder initialisation.
endMs	Turn end relative to recorder initialisation, or null.
durationMs	Captured or derived turn duration, or null.
text	STT result or the text supplied to TTS; null when unavailable or protected.
audio	Relative turn-audio paths keyed by raw and, when generated, mp3.
s3Audio	Full S3 object keys for the same audio.
mediaFormat	Raw encoding and rate, such as raw/lpcm16_16.
provider	STT or TTS provider identifier, or null.
model	Provider model or deployment identifier, or null.
stage	Source stage used for the segment.
nodeId	Matching flow node, or null.
storageRedacted	true when storage was suppressed for this turn.
redactionReason	sensitiveInfoOnStorage for a TOBi-protected turn.

Items remain in chronological order. Optional source metadata uses null; audio keys are omitted when that output was not generated.

Storage protection and retention

• The app Recording toggle controls whether audio and full transcript text are retained. There is no separate “transcripts only” switch.
• ENABLE_PII_LOGGING does not disable S3 transcript retention. It controls redaction in Postgres, SigNoz, and traces.
• Managed TOBi can set sensitiveInfoOnStorage for protected intervals. A protected turn remains in sequence but has no text or audio references and includes storageRedacted: true and redactionReason: "sensitiveInfoOnStorage".
• Team S3 lifecycle configuration determines how long the objects remain in the bucket.

See Conversations → Recording and transcript retention and TOBi managed LLM.

Operator prerequisites

1. Platform call recording is enabled for the deployment.
2. The team recording bucket, region, and role are configured under Settings → Team.
3. The app Recording toggle is enabled for flows that must retain audio and transcripts.

See also

• Conversations
• Team overview — Call recording
• Voice operations
• PII redaction