Recommended settings: Delta2

Delta2 is Palabra's simultaneous translation model. It translates speech while it is being recognized, without waiting for the end of a sentence, and voices the translation in short parts. Use the settings below to get started with Delta2.

Not available with Delta2 yet

• Glossaries are not supported.
• Voice timbre detection is not supported. Choose the voice with voice_id.
• Speaker diarization is not supported.

WebRTC settings

When publishing your audio track:

• DTX: disabled
• RED: disabled

Example with the LiveKit JS client:

await room.localParticipant.publishTrack(
    localTrack, { dtx: false, red: false, audioPreset: { maxBitrate: 32000, priority: "high" } }
);

Delta2 pipeline settings

The example translates Italian into English. Set source_language and target_language to your languages, see Supported languages. For a detailed description of each field, see the Translation settings breakdown or contact Palabra support.

{
  "input_stream": {
    "content_type": "audio",
    "source": {
      "type": "webrtc"
    }
  },
  "output_stream": {
    "content_type": "audio",
    "target": {
      "type": "webrtc"
    }
  },
  "pipeline": {
    "transcription": {
      "source_language": "it"
    },
    "translations": [
      {
        "target_language": "en",
        "translation_model": "delta2",
        "speech_generation": {
          "voice_cloning": false,
          "voice_id": "default_low"
        }
      }
    ],
    "translation_queue_configs": {
      "global": {
        "min_tempo": 0.75,
        "base_tempo": 1.2,
        "max_tempo": 1.6
      }
    },
    "allowed_message_types": [
      "translated_transcription",
      "partial_transcription",
      "partial_translated_transcription",
      "validated_transcription"
    ]
  }
}

How it differs from the default settings

Field	Value	What it does
translations[].translation_model	"delta2"	Translates this target language with Delta2. Set it for every target language of the task: Delta2 can't be mixed with other translation models. Speech recognition, speech generation and the queue of translated audio are set up for Delta2 automatically.
translation_queue_configs.global.base_tempo	1.2	Normal playback speed of the translated speech.
translation_queue_configs.global.min_tempo, max_tempo	0.75, 1.6	Playback slows down to min_tempo when little translated audio is queued, to avoid gaps, and speeds up to max_tempo when the queue grows.

Supported languages

Delta2 translates from any of these languages into any other of them. Source and target must be different languages.

Language	source_language	target_language
English	en	en
German	de	de
Spanish	es	es, es-eu
French	fr	fr, fr-eu
Italian	it	it
Dutch	nl	nl
Portuguese	pt	pt-la
Russian	ru	ru