PrakashPask/marathi-indic-speak-qlora
text-to-speech model by PrakashPask
Older snapshot: check the original source for current information.
From the publisher
Model card & documentation
Source previewRead the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.
This repository contains a completed QLoRA adaptation of bodhan-ai/indic-speak for Marathi text-to-speech using the Marathi configuration of ai4bharat/Rasa. The experiment trains LoRA adapters on audio-token prediction while keeping the 4-bit base model, SNAC codec, and Vocos decoder frozen. The official Rasa train split is used for optimization and the official test split is held out for evaluation. Loss measures teacher-forced audio-token prediction. It does not directly measure pronunciation, naturalness, speaker similarity, or intelligibility. The notebook: The immutable revisions and complete settings are recorded in runconfig.json. All 26,960 training rows and 2,995 test rows were successfully encoded. No gender, style, row-count, duration, or arbitrary token-length filter was applied. Both genders and all available styles were retained. Rasa has no speaker ID, so gender is only a proxy for conditioning: These labels represent gender groups, not verified individual speakers. The split-overlap audit found: Repeated transcripts do not mean the recordings are duplicated,…
Read the full model card ↗ · Preview checked 2026-11-05T21:32:40.891Z
Inside the original model card — Document outline
- Marathi Indic-Speak Adaptation with QLoRA
- Project links
- Results
- Method
- QLoRA configuration
- Dataset and conditioning
- Audio fertility
- Repository contents
- Reproducing the run
- Loading the adapter
- Generated comparisons
- Limitations
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Links from the model card
References supplied by the publisher, not independently verified endorsements. Check the destination before downloading files or entering credentials.
Documentation belongs to its respective authors. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
Model overview
PrakashPask/marathi-indic-speak-qlora is a text-to-speech model repository published by PrakashPask on Hugging Face. The source reports the peft library.
This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.
Model facts
- Task
- text-to-speech
- Library
- peft
- Recent downloads (30 days)
- 0
- Cumulative likes
- 0
- Hugging Face trending score
- 0
- Reported safetensors parameters
- Not reported
- Architecture
- Not reported
- License
- Not reported — check the model card
- Access
- Not gated by Hugging Face
- Created
- 2026-09-24T06:12:09.000Z
- Last modified
- 2026-09-24T06:32:31.000Z
Compatibility and lineage
Reported languages: mr
Reported base models: bodhan-ai/indic-speak
Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.
Use and evaluate
Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.
No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.
Source and freshness
Source: Hugging Face Hub. Metadata observed 2026-09-24T06:37:32.927Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.