---

language: [te]

license: other

tags: [text-to-speech, tts, orpheus, telugu, qlora]

pipeline_tag: text-to-speech

---

Orpheus Telugu Female (QLoRA v1) -- GGUF

A Telugu text-to-speech voice, finetuned from Orpheus. Voice prefix:

telugu_female.

Eval loss 3.1809, matching the male run's 3.1779. Trained on 19,859 clips

(~48 h), of which only 4 were dropped -- the benefit of a corpus with

ground-truth transcripts rather than ASR labels.

Measured median F0 of generated speech: 218.5 Hz, against the male model's

136.9 Hz with no overlap between the two sets. Voice identity was verified by

measurement rather than assumed, because a generic prefix on an earlier Tamil

model silently drifted toward the wrong gender.

See also the matching male voice

and the GGUF build.

Model details

| | |

|---|---|

| Base | canopylabs/3b-hi-pretrain-research_release (Orpheus, Llama-3.2-3B backbone) |

| Method | QLoRA (4-bit nf4 base, bf16 compute), r=64 alpha=128, attention + MLP; embeddings frozen |

| Schedule | 2 epochs, lr 2e-4 cosine, batch 1 x grad-accum 16 |

| Hardware | one 16 GB RTX 5060 Ti, trained locally |

| Corpus | SYSPIN Telugu (arpit-tiwari/syspin-telugu-tts) -- studio read speech with ground-truth transcripts |

How it works

Orpheus emits SNAC audio-codec tokens which a SNAC vocoder decodes to 24 kHz

mono PCM. Prompt format is {voice}: {text}; the voice is a literal text

prefix learned at training time, so this model has no zero-shot cloning and

takes no audio input.

Audio token layout: control tokens 128257-128262, audio offset 128266,

7 tokens per frame each offset by (position_in_frame 0-6) * 4096.

Inference

Recommended sampling: temperature=0.3, repetition_penalty=1.3.

Output is quiet (peaks ~0.2-0.4) -- peak-normalise or level it downstream.

Licence and provenance

Derived from the SYSPIN Telugu corpus. The mirror this was trained from does

not declare a licence; anyone intending commercial use should confirm terms

with the upstream corpus holders (IISc SPIRE Lab / Bhashini) rather than

relying on this repo.

Files

| Quantisation | Size | Notes |

|---|---|---|

| f16 | 6.2 GB | conversion source |

| Q8_0 | 3.3 GB | served in production |

| Q4_K_M | 2.0 GB | measured faster than Q8_0 (RTF 0.90 vs 1.09) at no audible cost |

| Q2_K | 1.4 GB | included for comparison only |

Runs with llama.cpp / llama-cpp-python. You still need a SNAC vocoder

(hubertsiuzdak/snac_24khz) to turn the emitted codec tokens into audio --

the GGUF alone produces tokens, not waveforms.