research-style comparison table, exact params, drop UTMOS
Browse files- README.md +10 -11
- comparison.png +2 -2
README.md
CHANGED
|
@@ -36,14 +36,14 @@ general speaker capacity is what makes <100M reachable for code-switch speech.
|
|
| 36 |
|
| 37 |

|
| 38 |
|
| 39 |
-
| Model | Params | Accent ↑ |
|
| 40 |
-
|---|---|---|---|---|
|
| 41 |
-
|
|
| 42 |
-
|
|
| 43 |
-
| **
|
| 44 |
-
| Kokoro-82M
|
| 45 |
-
|
| 46 |
-
\* Kokoro's accent (English-word recall)
|
| 47 |
\*\* Kokoro uses its own single voice (no target-voice cloning) and is not code-switch tuned, so SECS does
|
| 48 |
not apply. The point of this row: a generic 82M TTS does not deliver fixed-voice Hindi-English code-switch;
|
| 49 |
this 90M model does, at the 265M teacher's quality.
|
|
@@ -53,12 +53,11 @@ this 90M model does, at the 265M teacher's quality.
|
|
| 53 |
| axis | this 90M | 265M teacher | delta | 95% CI |
|
| 54 |
|---|---|---|---|---|
|
| 55 |
| code-switch accent | 0.820 | 0.831 | -0.011 | [-0.038, +0.016] |
|
| 56 |
-
| naturalness (UTMOS) | 3.259 | 3.224 | +0.034 | [-0.021, +0.089] |
|
| 57 |
| voice fidelity (SECS) | 0.851 | 0.860 | -0.009 | [-0.014, -0.003] |
|
| 58 |
| runaway-tail rate | 4.4% | 6.7% | | |
|
| 59 |
|
| 60 |
-
Accent is statistically even (delta -0.011)
|
| 61 |
-
|
| 62 |
|
| 63 |
## How it was built
|
| 64 |
|
|
|
|
| 36 |
|
| 37 |

|
| 38 |
|
| 39 |
+
| Model | Params | Accent ↑ | SECS ↑ | Tail ↓ |
|
| 40 |
+
|---|---|---|---|---|
|
| 41 |
+
| XTTS-Hinglish-443M | 443M | 0.861 | 0.855 | 4.9% |
|
| 42 |
+
| XTTS-Hinglish-265M | 265M | 0.831 | 0.860 | 6.7% |
|
| 43 |
+
| **XTTS-Hinglish-90M (this)** | **89.96M** | **0.820** | **0.851** | **4.4%** |
|
| 44 |
+
| Kokoro-82M | 82M | 0.886\* | n/a\*\* | n/a |
|
| 45 |
+
|
| 46 |
+
\* Kokoro's accent (English-word recall) is favoured by its English-primary design.
|
| 47 |
\*\* Kokoro uses its own single voice (no target-voice cloning) and is not code-switch tuned, so SECS does
|
| 48 |
not apply. The point of this row: a generic 82M TTS does not deliver fixed-voice Hindi-English code-switch;
|
| 49 |
this 90M model does, at the 265M teacher's quality.
|
|
|
|
| 53 |
| axis | this 90M | 265M teacher | delta | 95% CI |
|
| 54 |
|---|---|---|---|---|
|
| 55 |
| code-switch accent | 0.820 | 0.831 | -0.011 | [-0.038, +0.016] |
|
|
|
|
| 56 |
| voice fidelity (SECS) | 0.851 | 0.860 | -0.009 | [-0.014, -0.003] |
|
| 57 |
| runaway-tail rate | 4.4% | 6.7% | | |
|
| 58 |
|
| 59 |
+
Accent is statistically even (delta -0.011) and SECS passes non-inferiority, with a lower failure tail than
|
| 60 |
+
the teacher. A 3x smaller model at the same code-switch quality.
|
| 61 |
|
| 62 |
## How it was built
|
| 63 |
|
comparison.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|