Use and adapt Habibi-TTS

Use the documented interfaces and source layout for Habibi-TTS. The sections below retain the README’s examples and configuration details.

CLI Usage

1# Default using the Unified model (recommanded)
2habibi-tts_infer-cli \
3--ref_audio "assets/MSA.mp3" \
4--ref_text "كان اللعيب حاضرًا في العديد من الأنشطة والفعاليات المرتبطة بكأس العالم، مما سمح للجماهير بالتفاعل معه والتقاط الصور التذكارية." \
5--gen_text "أهلًا، يبدو أن هناك بعض التعقيدات، لكن لا تقلق، سأرشدك بطريقة سلسة وواضحة خطوة بخطوة."
6 
7# Assign the dialect ID, rather than inferred from given reference prompt (UNK, by default)
8# (best use matched dialectal content with ID: MSA, SAU, UAE, ALG, IRQ, EGY, MAR, OMN, TUN, LEV, SDN, LBY)
9habibi-tts_infer-cli --dialect MSA
10 
11# Alternatively, use `.toml` file to config, see `src/habibi_tts/infer/example.toml`
12habibi-tts_infer-cli -c YOUR_CUSTOM.toml
13 
14# Check more CLI features with
15habibi-tts_infer-cli --help

[!NOTE]
Some dialectal audio samples are provided under src/habibi_tts/assets, see the relevant README.md for usage and more details.

Training & Finetuning

See https://github.com/SWivid/Habibi-TTS/issues/2.

Benchmarking

0. Benchmark setup

1# Example template for benchmark use:
2python src/habibi_tts/eval/0_benchmark.py -d MSA
3# --dialect DIALECT (MSA | SAU | UAE | ALG | IRQ | EGY | MAR)

1. Generate benchmark samples with Habibi or 11Labs

1# Zero-shot TTS performance evaluation:
2accelerate launch src/habibi_tts/eval/1_infer_habibi.py -m Unified -d MAR
3# --model MODEL (Unified | Specialized)
4# --dialect DIALECT (MSA | SAU | UAE | ALG | IRQ | EGY | MAR)
5 
6# Use single prompt, to compare with 11Labs model:
7accelerate launch src/habibi_tts/eval/1_infer_habibi.py -m Specialized -d IRQ -s
8# --single (<- add this flag)
9 
10# Use single prompt, call ElevenLabs Eleven v3 (alpha) API:
11pip install elevenlabs
12python src/habibi_tts/eval/1_infer_11labs.py -a YOUR_API_KEY -d MSA
13# --api-key API_KEY (your 11labs account API key)
14# --dialect DIALECT (MSA | SAU | UAE | ALG | IRQ | EGY | MAR)

2. Transcribe samples with ASR models and calculate WER

1# Evaluate WER-O with Meta Omnilingual-ASR-LLM-7B v1:
2pip install omnilingual-asr
3python src/habibi_tts/eval/2_cal_wer-o.py -w results/Habibi/IRQ_Specialized_single -d IRQ
4# --wav-dir WAV_DIR (the folder of generated samples)
5# --dialect DIALECT (MSA | SAU | UAE | ALG | IRQ | EGY | MAR)
6# --batch-size BATCH_SIZE (set smaller if OOM, default 64)
7 
8# Evaluate WER-S with dialect-specific ASR models:
9python src/habibi_tts/eval/2_cal_wer-s.py -w results/Habibi/MAR_Unified -d MAR
10# --wav-dir WAV_DIR (the folder of generated samples)
11# --dialect DIALECT (EGY | MAR)

3. Calculate speaker similarity (SIM) between generated and prompt

Download WavLM Model from Google Drive, then

1python src/habibi_tts/eval/3_cal_spksim.py -w results/Habibi/MAR_Unified -d MAR -c YOUR_WAVLM_PATH
2# --wav-dir WAV_DIR (the folder of generated samples)
3# --dialect DIALECT (MSA | SAU | UAE | ALG | IRQ | EGY | MAR)
4# --ckpt CKPT (the path of download WavLM model)
5 
6python src/habibi_tts/eval/3_cal_spksim.py -w results/Habibi/IRQ_Specialized_single -d IRQ -c YOUR_WAVLM_PATH -s
7# --single (if eval single prompt or 11labs results)

4. Calculate UTMOS of generated samples

1python src/habibi_tts/eval/4_cal_utmos.py -w results/11Labs_3a/MSA
2# --wav-dir WAV_DIR (the folder of generated samples)

[!NOTE]
If conflicts after omnilingual-asr installation, e.g. flash-attn, try re-install
pip uninstall -y flash-attn && pip install flash-attn --no-build-isolation

Troubleshoot a local change

Reproduce the smallest example from the quick-start guide.
Compare required configuration and dependency versions with the README.
Check the linked issue tracker for the same error. Include the command, runtime version, and relevant error when reporting a problem; omit credentials.

Source and help

The catalog identifies the license as MIT. Read the repository license before redistributing source or assets.

This catalog entry is a fork. The README may describe upstream packages, domains, or release procedures; those destinations do not establish a separate release of this fork.