Audio modelOpen weights
Index-Echo-S2TT-9B
Bilibili Index Team
- Released
- -
- Data date
- October 7, 2026
Index-Echo-S2TT-9B uses a larger text decoder than the 2B variant, while retaining the same subtitle output. The source transcript and translation appear as timestamped text segments for Chinese to English, Japanese, or Spanish.
Specifications and access
| Specification | Value and source |
|---|---|
| Model type | Open weightsSource |
| Model ID | IndexTeam/Index-Echo-S2TT-9BSource |
| Model task | Speech translation into timestamped subtitlesSource |
| Access | Public Hugging Face weights for local executionSource |
| Input modalities | Audio or video containing speechSource |
| Output modalities | Transcript and translated subtitles with timestampsSource |
| License | Apache 2.0 according to the Hugging Face model cardSource |
| Translation directions | Chinese to English, Japanese, or SpanishSource |
| Decoder size | 9 billion parameters in the text decoder; not the total audio-package sizeSource |
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Index-Echo S2TT in-house video translation: MT judge (Show measurement, test conditions, and source)
- Source value
- 0.857
- Score
- 0.857
- Metric
- Gemini-3.1-Pro per-line translation quality rating on 0 / 0.5 / 1 scale
- Unit
- ratio
- Benchmark version
- arXiv:2609.40181v1, Table 7
- Category
- audio
- Direction
- Higher is better
- Sample
- 140
- Source type
- vendor-reported
- Evaluator
- Index-Translate paper v1, Table 7
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Index-Echo-S2TT-9B text decoder; 140 in-house video-translation windows of 50–60 seconds, 20 in each of seven directions (Chinese to English/Japanese/Korean/Spanish/Portuguese/Arabic, plus English to Chinese).
- Context
- Model-team in-house test; no external replication or per-direction sample breakdown is provided in the table row.
Index-Echo S2TT in-house video translation: median ASR error (Show measurement, test conditions, and source)
- Source value
- 0.102
- Score
- 0.102
- Metric
- median rate of source-audio transcript errors versus reference
- Unit
- ratio
- Benchmark version
- arXiv:2609.40181v1, Table 7
- Category
- audio
- Direction
- Lower is better
- Sample
- 140
- Source type
- vendor-reported
- Evaluator
- Index-Translate paper v1, Table 7
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Index-Echo-S2TT-9B text decoder; 140 in-house video-translation windows of 50–60 seconds, 20 in each of seven directions (Chinese to English/Japanese/Korean/Spanish/Portuguese/Arabic, plus English to Chinese).
- Context
- Model-team in-house test; no external replication or per-direction sample breakdown is provided in the table row.
Index-Echo S2TT in-house video translation: sentence start-time MAE (Show measurement, test conditions, and source)
- Source value
- 0.488
- Score
- 0.488
- Metric
- mean absolute sentence start-time error
- Unit
- seconds
- Benchmark version
- arXiv:2609.40181v1, Table 7
- Category
- audio
- Direction
- Lower is better
- Sample
- 140
- Source type
- vendor-reported
- Evaluator
- Index-Translate paper v1, Table 7
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Index-Echo-S2TT-9B text decoder; 140 in-house video-translation windows of 50–60 seconds, 20 in each of seven directions (Chinese to English/Japanese/Korean/Spanish/Portuguese/Arabic, plus English to Chinese).
- Context
- Model-team in-house test; no external replication or per-direction sample breakdown is provided in the table row.
Sources and data date
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 7, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | independent-s2tt-replication Original wording The S2TT scores come from the model team's in-house video test; no independent repetition for this checkpoint was found in the sources read. |
| Additional source | Index-Echo-S2TT-9B Hugging Face model card (retrieved October 7, 2026) · Index-Echo-S2TT-9B model card · Editorial description reviewed October 7, 2026 |
| Additional source | Bilibili Index-Translate repository (retrieved October 7, 2026) · Bilibili Index-Translate repository · Editorial description reviewed October 7, 2026 |
| Additional source | Index-Translate paper v1, Table 7 (retrieved October 7, 2026) |