Lem (澪夢レム)
UTAU & Diffsinger Voicebank & Furry Character
Lem is a shapeshifter character;
a voice for various voice synthesis softwares.
Lem the Shapeshifter, also known as Lem.ma or 澪夢レム, is a free virtual singer and AI vocal synthesizer similar to Vocaloid or Synthesizer V. Powered by the open-source engines OpenUTAU and Diffsinger, Lem UTAU voicebanks provide multipitch vocal libraries with an exhausitve phonetic set and rigorous diphone coverage, ensuring seamless, high-fidelity synthesis across multiple languages. For modern AI workflows, a free Diffsinger model is also available, bringing that same linguistic depth to neural rendering.
Beyond the voice, Lem is a character designed for adaptation. His visual identity serves as a creative framework for creation of derivative lem-shapes. Whether utilizing his Civet, Quoll, or Phascogale forms, or deriving entirely new iterations, Lem invites you to reshape his design to fit your unique artistic vision.
Reference sheets (MEGA) Character information Visual Protocol TOS / Terms of Use
Voice Showcase
A selection of works demonstrating Lem in various musical contexts. Ranging from original songs to community covers.
Downloads
Lem V4Bi: English Arpasing + Japanese
澪夢レムV4★英語アーパシング + 日本語
Voicebank Terms of Use
(read & scroll for download link)
By downloading and or publishing any work that is wholly or partially made with this software you agree to comply with the Terms Of Use.
All V4 English Voicebanks (MEGA) Lem V4 Civet English (MEGA) Lem V4 Phascogale English (MEGA) Lem V4 Quoll English (MEGA)
Looking for a free, highly capable English UTAU download? Lem V4 Arpasing brings the expressive power of a furry vocaloid alternative to your music production. Designed for OpenUTAU, this voicebank features comprehensive Arpasing phoneme coverage alongside extended vowels and consonants for unparalleled tuning control and multilingual capabilities.
While meticulously recorded for English and Japanese, Lem V4 transcends language barriers. Thanks to robust dictionary suffix support, Lem can sing in the following extra languages: Korean, Chinese, Cantonese, French, and Spanish. To unlock these extra languages, please refer to the OpenUTAU Yaml Dictionaries multilingual support tutorial.
| Voicebank (EN) | Composition | Data Length | |
|---|---|---|---|
| Civet (シベット) | 5 Main + 5 Head Voice A2 to A3 Main A3 to A4 Head Voice |
1.50 GB | 11 hours, 36 minutes |
| Quoll (フクロネコ) | 5 Main + 3 Head Voice A2 to A3 Main C#4 to A4 Head Voice |
1.24 GB | 9 hours, 17 minutes |
| Phascogale (ファスコガーレ) | 5 Main + 3 Head Voice A2 to A3 Main C#4 to A4 Head Voice |
1.24 GB | 9 hours, 17 minutes |
Technical Specifications (EN)
Asaxi Conlang Phonemizer (Coming soon)
nn, nng, mm, xn,
a, i, u, e, o
cl, exh, inh, sil, q,
ky, gy, ngy, ty, dy, ny, my, hy, fy, by, py, my, dxy, vy, ly, ry
Installation & Usage Instructions
1. Download the voicebank .zip file from the links above.
2. Locate your OpenUTAU
Singers directory. (In OpenUTAU, "Select Singer" > Open singers Location)3. Place the downloaded .zip file directly into the
Singers folder.4. Unzip and delete the .zip file.
5. In OpenUTAU, click "Refresh" in the "Select Singer" menu. The new voicebank should now appear in the list.
Important Usage Notes:
• Normalization: Change the default normalize (norm) parameter from 86 to 0. This must be done in every USTx project separately by clicking the cog icon on the bottom right of the piano-roll. This applies to all tracks. Synthesized output is normalized too intensely by default. Changing it allows for better mixing in post because the output loudness will have natural dynamics.
• Phonemizers: Intended for use in OpenUTAU with the EN ARPA + phonemizer. Can also be used with the Asaxi language phonemizer.
• Resamplers: Tested and verified on TIPS.exe, worldline-r (default), doppeltler64.exe, and hifisampler.exe.
• Hifisampler note: Normalization is turned off in its configuration yaml by default. Make sure the
mod+ parameter is above 0 to avoid bad phase alignment in transitions.
Lem V4 Japanese ★ CVVC
澪夢レムV4
Voicebank Terms of Use
(read & scroll for download link)
By downloading and or publishing any work that is wholly or partially made with this software you agree to comply with the Terms Of Use.
All V4 Japanese Voicebanks (MEGA) Lem V4 Civet Japanese (Bowlroll) Lem V4 Quoll Japanese (Bowlroll) Lem V4 Phascogale Japanese (Bowlroll)
Lem V4 Japanese is a series of three unique voicebanks, each recorded in a different larynx position. Each "Voice Colour" is a standalone library containing two distinct vocal modes: Main and Headvoice.
| Voicebank (JP) | Composition | Weight | Recording Stats |
|---|---|---|---|
| Civet (シベット) | 5 Main + 5 Head Voice A2 to A3 Main A3 to A4 Head Voice |
397 MB |
Time: 3h 18m 30.12.2024 - 17.08.2025 |
| Quoll (フクロネコ) | 5 Main + 3 Head Voice A2 to A3 Main C#4 to A4 Head Voice |
316 MB |
Time: 2h 27m 01.10.2025 - 19.03.2025 |
| Phascogale (ファスコガーレ) | 5 Main + 3 Head Voice A2 to A3 Main C#4 to A4 Head Voice |
361 MB |
Time: 2h 27m 24.03.2025 - 28.09.2025 |
Technical Specifications (JP)
Lem VCCV / CVVC Mongoose
澪夢レムCVVC★マングース
Voicebank Terms of Use
(read & scroll for download link)
By downloading and or publishing any work that is wholly or partially made with this software you agree to comply with the Terms Of Use.
Lem English VCCV / JP CVVC Mongoose (MEGA) Lem English VCCV / JP CVVC Mongoose (Google Drive)
Lem English VCCV / JP CVVC 2024 is a bilingual English and Japanese voicebank. The illustration included with the voicebanks depicts Lem as a mongoose.
| Append | Pitches | Weight | Recording Stats |
|---|---|---|---|
| Normal | 5 Pitches A2 to A3 |
303 MB |
Time: 0h 50m 30.06.2024 - 05.11.2024 |
| Soft | 5 Pitches A3 to A4 (Head Voice) |
303 MB |
Time: 0h 50m 17.07.2024 - 18.11.2024 |
Technical Specifications
Lem V3 Weasel
澪夢レムV3★イイズナ
Voicebank Terms of Use
(read & scroll for download link)
By downloading and or publishing any work that is wholly or partially made with this software you agree to comply with the Terms Of Use.
Lem V3 Weasel is a VCV UTAU voicebank characterized by an outrageously bright, shimmering sound quality, due to the high larynx position and intense throat tension.
| Append | Pitches | Weight | Recording Stats |
|---|---|---|---|
| Main (Nasal) | 9 Pitches G2 to A4 |
605 MB |
Time: 1h 45m 1.02.2022 - 21.02.2022 |
Technical Specifications
Lem Diffsinger V2.4 "Marten"
澪夢レムDS★テン
Voicebank Terms of Use
(read & scroll for download link)
By downloading and or publishing any work that is wholly or partially made with this software you agree to comply with the Terms Of Use.
Lem Diffsinger V2.4 MM (Megamodel) is a diffsinger voice model trained in Pix's multispeaker Megamodel. Characterized by a very dynamic vocal whose tone varies from phrase to phrase at random due to a lack of distinct voice colours.
| Languages | Range (Optimal) | Parameters | Weight / Info | Data Length English | Data Length Polish |
|---|---|---|---|---|---|
|
Japanese English Polish |
A2 to E5 (Sensitive to pitch) |
TENC (Tension) VELC (Speed) GENC (Gender) |
360 MB (whole zip) 259 MB (onnx) |
54 minutes approx. 02.03.2024 - 05.05.2024 |
15 minutes approx. recorded on 18.05.2023 |