32 short recordings (Jordanian, Palestinian, Qatari and Saudi Arabic, and Hindi) used as reference voices for voice cloning with VoxCPM-Kernel/VoxCPM2-Arabic-Dialects. They are inputs, not model outputs: the engine's latency benchmark clones every one of them in the dialect adapter its file name names.

At a glance


What	32 reference clips, WAV, mono, 16-bit PCM
Total length	242 s (3-15 s per clip)
Sample rates	16, 24 or 48 kHz (as supplied)
Transcripts, speaker metadata	none supplied
Used by	`examples/benchmark.py` and `tests/gpu/verify_engine_fixes.py` of Speech-Next/voxcpm2-arabic-dialects-turbo
Access	gated: request access on this page, then `hf auth login`
License	not stated by the source

language	clips	length	adapter
Jordanian Arabic	17	140.0 s	`jordanian`
Palestinian Arabic	5	23.7 s	`palestinian`
Qatari Arabic	3	16.5 s	`qatari`
Saudi Arabic	3	29.0 s	`saudi`
Hindi	4	33.3 s	`hindi_english`

Files

The file name is <Dialect><Name>[suffix].wav. The speaker column is the name in the file name; sex is given only where the file name says it (_F, _M).

file	language	adapter	speaker	sex	sample rate	length
`HindiPooja_F.wav`	Hindi	`hindi_english`	Pooja	female	16 kHz	7.5 s
`HindiRahul_M.wav`	Hindi	`hindi_english`	Rahul	male	16 kHz	3.3 s
`HindiRohan_M.wav`	Hindi	`hindi_english`	Rohan	male	16 kHz	14.6 s
`HindiSimran_F.wav`	Hindi	`hindi_english`	Simran	female	16 kHz	7.9 s
`JordanianAnas.wav`	Jordanian Arabic	`jordanian`	Anas		24 kHz	12.0 s
`JordanianAsmaa.wav`	Jordanian Arabic	`jordanian`	Asmaa		48 kHz	5.9 s
`JordanianAsmaa_0.wav`	Jordanian Arabic	`jordanian`	Asmaa		24 kHz	12.4 s
`JordanianDana.wav`	Jordanian Arabic	`jordanian`	Dana		24 kHz	4.5 s
`JordanianFarah.wav`	Jordanian Arabic	`jordanian`	Farah		48 kHz	10.6 s
`JordanianLaith.wav`	Jordanian Arabic	`jordanian`	Laith		24 kHz	10.6 s
`JordanianMalak.wav`	Jordanian Arabic	`jordanian`	Malak		48 kHz	13.3 s
`JordanianMalak_0.wav`	Jordanian Arabic	`jordanian`	Malak		48 kHz	12.5 s
`JordanianMalak_2.wav`	Jordanian Arabic	`jordanian`	Malak		48 kHz	7.7 s
`JordanianMisk.wav`	Jordanian Arabic	`jordanian`	Misk		16 kHz	4.8 s
`JordanianMohannad.wav`	Jordanian Arabic	`jordanian`	Mohannad		16 kHz	3.1 s
`JordanianQusai.wav`	Jordanian Arabic	`jordanian`	Qusai		16 kHz	7.2 s
`JordanianRasheed.wav`	Jordanian Arabic	`jordanian`	Rasheed		16 kHz	5.8 s
`JordanianRawan.wav`	Jordanian Arabic	`jordanian`	Rawan		24 kHz	4.4 s
`JordanianSarah.wav`	Jordanian Arabic	`jordanian`	Sarah		48 kHz	8.5 s
`JordanianTala.wav`	Jordanian Arabic	`jordanian`	Tala		24 kHz	3.6 s
`JordanianZaid2.wav`	Jordanian Arabic	`jordanian`	Zaid		24 kHz	13.0 s
`PalestinianAhmad.wav`	Palestinian Arabic	`palestinian`	Ahmad		24 kHz	5.1 s
`PalestinianKhalid.wav`	Palestinian Arabic	`palestinian`	Khalid		24 kHz	4.7 s
`PalestinianMosab.wav`	Palestinian Arabic	`palestinian`	Mosab		24 kHz	5.8 s
`PalestinianRaghad1.wav`	Palestinian Arabic	`palestinian`	Raghad		24 kHz	3.9 s
`PalestinianTala1.wav`	Palestinian Arabic	`palestinian`	Tala		24 kHz	4.2 s
`QatariJassim.wav`	Qatari Arabic	`qatari`	Jassim		24 kHz	5.3 s
`QatariSuhail.wav`	Qatari Arabic	`qatari`	Suhail		24 kHz	6.1 s
`QatariTamim.wav`	Qatari Arabic	`qatari`	Tamim		24 kHz	5.1 s
`SaudiReem_38.wav`	Saudi Arabic	`saudi`	Reem		24 kHz	7.5 s
`SaudiSalman_38.wav`	Saudi Arabic	`saudi`	Salman		24 kHz	9.1 s
`SaudiSara_38.wav`	Saudi Arabic	`saudi`	Sara		24 kHz	12.4 s

Use

import os
from huggingface_hub import snapshot_download

folder = snapshot_download("VoxCPM-Kernel/Arabic-Hindi-Reference-Voices", repo_type="dataset", allow_patterns=["*.wav"])
voices = sorted(f for f in os.listdir(folder) if f.endswith(".wav"))

With the engine: python examples/benchmark.py --voices VoxCPM-Kernel/Arabic-Hindi-Reference-Voices (the default).

Source and license

The 32 files come unchanged, with their original names, from an archive supplied for this project. It held no transcripts, speaker metadata or license information.

Downloads last month: 16

Total file size:

13.7 MB