Speech → Magic Voice
Colours
Classic blue
CRT green
CRT amber
Black & white
Style
Normal
CRT
Open project
Save project
1 · Preview
Import audio file
New word
Edit existing word
Original
Zoom in
Zoom out
Zoom to selection
Show all
▶
■
Drag the
bright cursors
to choose the part to convert: the selected part is shown bright, the rest dimmed.
Settings
Voicing threshold
Lower = more buzzy (voiced) frames, higher = more hissy.
Level (dB)
Loudest sample relative to full scale. The selection is always normalised to full scale before it is converted (so the 7-bit energy range is used to the full); this setting then lowers the output from there.
Silence gate (dB)
Frames this far below loud speech become silent frames. Below about -42 dB the chip's lowest energy step is silence anyway, so lower settings change little.
Pitch range (Hz)
Min and max fundamental.
Monotone pitch (Hz)
0 = natural pitch. Any value gives the classic robot voice.
Silence
Compact silence
True silence
A real silent frame costs 96 bits (600 bytes/s), as much as a voiced one. Compact silence stores pauses as the cheapest frame there is (52 bits, 325 bytes/s) at the lowest energy: a hiss about 50 dB below full scale, practically inaudible, and about 45% less data in the pauses. True silence uses real silent frames: no hiss, but 600 bytes/s. This setting belongs to each word, so you can keep the same sample in both variants.
Soften P, B, F, S, T
Off
Mild
Strong
Bursts and hissing sounds (P, B, T, F, S, "st") can come out of the synthesiser louder and sharper than in the recording, and can also smear into the pause before them. This setting checks every such frame against the recording and turns it down where it is too loud, drops frames that are only smear, and caps the hiss at a fraction of the vowel level. Mild only corrects what is clearly too much; Strong also shaves the peaks. Off leaves the analysis as it is. It belongs to each word.
Compressor
Off
Mild
Medium
Strong
Makes soft passages louder and loud passages softer before the conversion, and then scales the whole recording to the maximum level, so that the 7-bit energy range of the chip is used as well as possible (best signal to noise ratio). Pauses and background noise are not amplified. Mild is gentle (ratio 1.6:1), Medium suits most speech (2.5:1), Strong evens out everything (4:1). This setting belongs to each word. It changes only what is converted: the original always plays unchanged.
Autosplit
Enabled
Disabled
When the selection is too long for the memory of one vocabulary: Enabled cuts it at pauses into several words (NAME000, NAME001, …) and spreads them over as many vocabularies as needed; the cuts are shown in the converted graph. Disabled flags a selection that is too long and leaves it out of the export.
Reset to defaults
Convert
Converted
▶
■
original (level-matched)
Magic Voice output · click the graph to start from that point
Details
Word name
Suggest name
Add to vocabulary gallery
Save changes
2 · Vocabulary gallery
Load address
3 · Export
Information
Import a recording, convert it and add words to the vocabulary gallery.