Add README.md
Browse files
README.md
CHANGED
|
@@ -35,9 +35,16 @@ IB-Robot computes these on the host in
|
|
| 35 |
`src/voice_asr_service/voice_asr_service/speaker_id/features.py`, which is pinned
|
| 36 |
against sherpa-onnx's own extractor by `test_speaker_features.py`.
|
| 37 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
| Tensor | Semantic | Shape | dtype |
|
| 39 |
|---|---|---|---|
|
| 40 |
-
| input | `host.campplus.features` | `(1,
|
| 41 |
| output | `host.campplus.embedding` | `(1, 192)` | float32 |
|
| 42 |
|
| 43 |
## Deployments
|
|
|
|
| 35 |
`src/voice_asr_service/voice_asr_service/speaker_id/features.py`, which is pinned
|
| 36 |
against sherpa-onnx's own extractor by `test_speaker_features.py`.
|
| 37 |
|
| 38 |
+
The host also splits an utterance into **fixed 200-frame (2.0 s) windows**, which
|
| 39 |
+
is what the input shape above reflects. Shorter utterances are tiled from their
|
| 40 |
+
own frames rather than zero-padded; longer ones become consecutive windows with
|
| 41 |
+
the last aligned to the end, and the per-window unit vectors are averaged. This
|
| 42 |
+
is part of the declared preprocessing contract, so every deployment applies it
|
| 43 |
+
and vectors stay comparable across them.
|
| 44 |
+
|
| 45 |
| Tensor | Semantic | Shape | dtype |
|
| 46 |
|---|---|---|---|
|
| 47 |
+
| input | `host.campplus.features` | `(1, 200, 80)` | float32 |
|
| 48 |
| output | `host.campplus.embedding` | `(1, 192)` | float32 |
|
| 49 |
|
| 50 |
## Deployments
|