Primary use stated or inferred cautiously from official documentation.
Phonon 31M English
Gradium text-to-speech model with 31 million parameters, excluding the audio tokenizer. Published on October 19, it added its Apache-2.0 license file on November 17. It generates English speech with 64.5 MB q8 GGUF weights.
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
Limits and integration
| API ID | gradium/phonon-31m-en |
|---|---|
| Model type | Audio |
| Access model | Downloadable weights |
| License | Apache-2.0 |
| Deployment | Self-hosted |
| Release | 19 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text |
| Exit | Audio |
| Context window | English text; maximum length unpublished |
| maximum output | Mono audio at 24 kHz, generated in streaming |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face |
Local speech synthesis in a browser, phone or CPU. It uses a flow-matching transformer and the Mimi decoder to produce 80 ms audio frames. The license covers weights, tokenizer and voices in this repository.
This profile covers only the 31M English model. The 40M variants use other licenses and do not inherit Apache-2.0. Audio frames of 80 ms do not guarantee that latency on every device. Review the runtime and permissions for additional voices.
What it can do
Orientation
Local speech synthesis in a browser, phone or CPU. It uses a flow-matching transformer and the Mimi decoder to produce 80 ms audio frames. The license covers weights, tokenizer and voices in this repository.
Context
English text; maximum length unpublished · Mono audio at 24 kHz, generated in streaming
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
The repository publishes Apache-2.0, including commercial use under its terms. The runtime is separately licensed MIT or Apache-2.0. No official API price is verified; local execution consumes your own resources.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
How to read the results
The card reports English US Elo 2098 from 468 raters and 8,290 votes. This is a manufacturer listening-preference comparison, not an independent measurement or a pronunciation guarantee for every text.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| Phonon English US listening Elo | 2098 | Elo · listening preference | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
Chronology
gradium/phonon-31m-en
Original weights publication.
8f7a37c9
Apache-2.0 file added to the repository; the model’s original date is unchanged.