Singing Synthesis: VOCALOID and Synthesizer V

You draw notes and type words to get a singing voice; pitch is exact, diction and expression are hand-worked.

Model profile

Vendor

Yamaha · Dreamtonics

Form

本地模型

Input

MIDI · 歌词 · 音符

Output

人声 · 音频

Licence

声库与编辑器的商用条款、署名要求、声优同意范围需按各声库的 EULA 核对，录音与训练来源以厂商说明为准

Verified

2026-10-30

Related

VibratoTimbreVoice

What it lets you finish

A demo vocal with pitch you control completely; the same melody tried in another language; four-part harmony or a doubled lead from one project; several readings of one line to compare; a clean vocal stem handed to a mixer. It does not make backing tracks, and it does not turn a sentence into a song.

Inputs and outputs

You input notes and words: pitch and duration drawn on a piano roll, one syllable per note, then vibrato, dynamics, breathiness and breath points. MIDI import and melody extraction from audio are also available. The output is a vocal track, exported whole or in sections to cut against picture. The project stores parameters rather than audio, so swapping the voicebank gives the same project another singer.

Strengths

• Pitch and timing precise enough for ornaments, glides and leaps a singer would find hard.
• Changing one word or one note needs no retake; iteration is nearly free.
• Harmony, doubles and language versions derive from one project and update together.
• The output is a dry stem with a low noise floor, easy to mix and process.
• Good for a listenable demo before deciding whether to book a singer.

Weaknesses and failure modes

• Diction: consonant clusters, liaisons and Chinese polyphonic characters must be set by hand, or the voice sings the wrong word, smoothly enough to pass a casual listen.
• Expression: without drawn parameters, long-note vibrato cycles too regularly and breaths land on the barline instead of the phrase, which reads as sight-reading.
• Fixed timbre: each voicebank has one basic colour and loses conviction in rock, rap or dramatic soprano material.
• Language edges: a Japanese voicebank singing Chinese, or a Chinese one singing English, distorts vowels and stress; where tone and melody conflict, one of them has to change.
• Splicing: next to a real recording the noise floor and space do not match, so room tone or a matching chain is needed.
• Performance: ornaments and breaths are drawn, not felt, and used heavily the result is too even to believe.

Terms, price and alternatives

Questions to check: does the voicebank licence cover commercial release? Is credit to the voicebank or vendor required? How far does the character likeness licence reach? Does the voice provider's consent cover adaptation, translation and redistribution? May you hand the exported dry stem to a third party?

Pricing is mostly a one-off editor plus voicebanks, with some cloud services billed as subscriptions; regional availability is as stated on the official pages.

Alternatives: ACE Studio when you want fewer hand-set parameters and more natural Chinese diction; an open-source system such as DiffSinger when you need full offline work or a voice you trained; a session singer when you need improvisation, breath and real emotion, where synthesis still falls a layer short.

Sources

• VOCALOID 与 Synthesizer V 的编辑方式，依厂商公开说明核对，未抄原文
• 歌声合成的咬字与表情处理，依制作实践通识核对，未抄文本

Used for fact-checking only; all prose is original.