A local LatAm conversion failed the content gate
A local MLX conversion of the LatAm package was compared with the existing multilingual Q8 and an original FP32 route. All active weights were loaded strictly, but that alone did not establish correct speech.
Key finding
The deployed multilingual Q8 remained the choice: the local LatAm Q8 conversion added nine content edits over 142 words.
Edited finding from the pinned experiment record. The report and figure metadata preserve the original workload and limits.
The boundary
Five Spanish phrases, two seeds, one authorized private reference and concurrent desktop load. The upstream PyTorch runtime was not tested.
Data
Reviewed aggregate results with explicit units, sources and qualification limits. Different cohorts remain separate.
The local LatAm conversions added content edits
Five Spanish phrases · two seeds · matched local conversion test
- Current multilingual Q80
- Community FP320
- Local LatAm Q89
- Local LatAm FP328
- Q8 · FP16 vocoder trial0
edits / 142 reference words
Recorded content normalization differs from strict WER. Upstream PyTorch was not tested.
View data & source
| Configuration | Value |
|---|---|
| Current multilingual Q8 | 0 |
| Community FP32 | 0 |
| Local LatAm Q8 | 9 |
| Local LatAm FP32 | 8 |
| Q8 · FP16 vocoder trial | 0 |
Origin: reviewed report table. Revision 7e6fdbe3b1d1.
Write-ups
Report chatterbox-fidelity/community6-vs-coreml-2026-10-10/latam-mlx-test/README.md
Accent labels are not acceptance results
A local MLX conversion of the LatAm package was compared with the existing multilingual Q8 and an original FP32 route. All active weights were loaded strictly, but that alone did not establish correct speech.
What the paired clips showed
The current Q8 and community FP32 each had zero content edits over 142 reference words after the recorded normalization. Local LatAm Q8 had nine; local LatAm FP32 had eight. Those counts do not diagnose a defect in the upstream model because its original runtime was not evaluated.
An FP16-vocoder experiment preserved the Q8 output closely. A cleaner five-pair AB/BA recheck measured 2.969 versus 2.970 seconds, both RTF 0.675, with identical speech tokens. It established no useful speed gain.
Keep the working default
The tested conversion failed the content gate, so multilingual Q8 stayed selected. Strict and normalized word edits remain separate, and neither evaluates accent, voseo, emotion or human naturalness. No private reference recording is distributed here.
Evidence and publication boundary
This is an edited account of the recorded experiment. The linked archive is private; the methods, aggregate findings and limits are included above. Reference recordings, raw conversations, authentication material and workstation configuration are excluded.
More in Speech & codecs
MOSS on a 3090: Q2 failed, AWQ4 stayed closer
Every profile began with the pinned BF16 generator. The comparison reused fixed reference codes and an FP32 decoder for the main quality measurements, keeping codec changes separate from generator quantization.
AWQ4 stayed close to the BF16 reference on WavLM and character errors; the sensitivity Q2 candidate was rejected.
Drama3: BF16 decoded the same codes about three times faster
The experiment replayed saved Spanish and English codes representing 21.5481 and 15.8824 seconds of audio. It compared FP32, BF16, mixed BF16 and two INT8 decoder variants on one RTX 3090.
BF16 reduced median decoder time from 390.07 to 129.54 ms in Spanish and from 292.42 to 96.88 ms in English.
Keeping FP32 and moving Supertonic to Triton
Rather than reduce the CPU graph precision again, this follow-up used NVIDIA Triton’s ONNX backend with CUDA execution and the unchanged FP32 graphs. The paired comparison retained both speed and acoustic checks.
The paired GPU test cut median first-audio time to about 0.35 seconds while preserving the measured ASR results.