Audiobox TTA-RAG · retrieved text and audio / 3

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit

Not collected

Maximum output

Not collected

Input modalities

Text, AudioSource for Input modalities (opens in a new tab)

Output modalities

AudioSource for Output modalities (opens in a new tab)

Reasoning

Not collected

Tool calling

Not collected

Structured output

Not collected

Streaming

Not collected

Batch processing

Not collected

Fine-tuning

Not collected

Knowledge cutoff

Not collected

Release date

2024-11-07Source for Release date (opens in a new tab)

Research disclosure date, not general API or weight availability.

Parameters

Not collected

Architecture

Audiobox flow-matching generator with Encodec features, T5-base text encoder and text-conditioned retrieval-audio encoderSource for Architecture (opens in a new tab)

Languages

Not collected

Context caching

Not collected

No model-specific inference service or discounted prompt-cache contract is established for these research weights.

Published training data

AudioCaps official training split: more than 45,000 captioned audio examples, approximately ten seconds eachSource for Published training data (opens in a new tab)

retrievalEncoder

Three transformer layers, 16 heads, 1024-dimensional hidden state, 4096-dimensional feed-forward layerSource for retrievalEncoder (opens in a new tab)

trainingRetrieval

Three AudioCaps examples selected by text-to-text retrieval, with both audio and captionsSource for trainingRetrieval (opens in a new tab)

retrievalConditioning

Interleaved Encodec audio features and T5-base caption featuresSource for retrievalConditioning (opens in a new tab)

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

Why it is included

The Meta/University of Texas at Dallas authors publish the TTA-RAG paper and a linked audio-sample site. The reviewed artifacts distribute example audio, not the original baseline or retrieval-adapted checkpoints, and supply no callable API. Public CLAP, Encodec and T5 components are not the trained TTA-RAG system.

Lifecycle dates

Sources reviewed