Model card: per-window RTFs incl. lessac-low 1.5/3/5 s; refresh timings
Browse files
README.md
CHANGED
|
@@ -45,14 +45,27 @@ model load excluded), median of 5 runs:
|
|
| 45 |
|
| 46 |
| Voice | Audio | CPU only | CPU + NPU | Speedup | First audio (CPU / CPU + NPU) |
|
| 47 |
|---|---|---|---|---|---|
|
| 48 |
-
| `en_US-libritts_r-medium` | 7.70 s | 5.
|
| 49 |
-
| `en_US-lessac-low` | 9.82 s | 5.
|
| 50 |
-
| `es_MX-ald-medium` | 12.63 s | 8.
|
| 51 |
|
| 52 |
"CPU only" is the full Piper voice in onnxruntime (2 threads), one sentence at a
|
| 53 |
time; "CPU + NPU" is partA on the CPU pipelined with the vocoder on the NPU. The
|
| 54 |
English and Spanish samples are the same three sentences, translated.
|
| 55 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
The CPU+NPU pipeline overlaps the halves: the CPU encodes sentence *n+1* while the
|
| 57 |
NPU vocodes sentence *n* and the speaker plays sentence *n-1*.
|
| 58 |
|
|
|
|
| 45 |
|
| 46 |
| Voice | Audio | CPU only | CPU + NPU | Speedup | First audio (CPU / CPU + NPU) |
|
| 47 |
|---|---|---|---|---|---|
|
| 48 |
+
| `en_US-libritts_r-medium` | 7.70 s | 5.55 s (1.39× RT) | **2.83 s (2.72× RT)** | **1.96×** | 0.95 s / **0.77 s** |
|
| 49 |
+
| `en_US-lessac-low` | 9.82 s | 5.25 s (1.87× RT) | **2.51 s (3.92× RT)** | **2.09×** | 0.89 s / **0.68 s** |
|
| 50 |
+
| `es_MX-ald-medium` | 12.63 s | 8.66 s (1.46× RT) | **4.07 s (3.10× RT)** | **2.13×** | 1.81 s / **1.38 s** |
|
| 51 |
|
| 52 |
"CPU only" is the full Piper voice in onnxruntime (2 threads), one sentence at a
|
| 53 |
time; "CPU + NPU" is partA on the CPU pipelined with the vocoder on the NPU. The
|
| 54 |
English and Spanish samples are the same three sentences, translated.
|
| 55 |
|
| 56 |
+
The vocoder alone on the NPU, per window:
|
| 57 |
+
|
| 58 |
+
| window | libritts / Spanish (22.05 kHz) | lessac-low (16 kHz) |
|
| 59 |
+
|---|---|---|
|
| 60 |
+
| 1 s | 170 ms (5.9× RT) | 126 ms (7.8× RT) |
|
| 61 |
+
| 1.5 s | — | 186 ms (8.1× RT) |
|
| 62 |
+
| 2 s | 335 ms (6.0× RT) | 249 ms (8.0× RT) |
|
| 63 |
+
| 3 s | — | 365 ms (8.2× RT) |
|
| 64 |
+
| 4 s | 688 ms (5.8× RT) | 497 ms (8.0× RT) |
|
| 65 |
+
| 5 s | — | 628 ms (8.0× RT) |
|
| 66 |
+
| 6 s | 1032 ms (5.8× RT) | 749 ms (8.0× RT) |
|
| 67 |
+
| 8 s | 1356 ms (5.9× RT) | 1027 ms (7.8× RT) |
|
| 68 |
+
|
| 69 |
The CPU+NPU pipeline overlaps the halves: the CPU encodes sentence *n+1* while the
|
| 70 |
NPU vocodes sentence *n* and the speaker plays sentence *n-1*.
|
| 71 |
|