How to Make SAM TTS Sing: Try Singing Mode and Build a Four-Beat Chant
Try SAM singing mode in your browser, hear speech-versus-singing samples, and reproduce a four-beat robotic chant with exact pitch values and WAV downloads.
Start with La. Render one pitch at a time, then save each syllable as WAV.
Pronunciation helpNames & numbers
Words, acronyms, and numbers that often need phonetic spelling will appear here.
Download name: microsoft-sam-sam-tts-singing.wav
Batch line rendering1/12 lines
Put each take on its own line, then download separate WAV files with the current voice and settings.
In this guide
Try singing mode with one syllable first
The experiment above exposes sam-js's singing mode. Leave SAM singing mode checked, type La., and generate a take at Speed 72, Pitch 64, Mouth 128 and Throat 128. Uncheck the mode and render the same syllable again. The two recordings below reproduce that comparison so you can hear the difference before editing anything.
Singing mode removes one layer of the engine's speech pitch contour. It gives short voiced fragments a steadier character, but does not read a score, set a tempo or turn a whole sentence into a natural sung melody. The site runs Software Automatic Mouth, which is a different engine from the original Windows Microsoft Sam voice.
La.Speed 72 · Pitch 64 · Mouth 128 · Throat 128Speech mode
Reference syllable with ordinary speech pitch shaping. Compare with the middle-pitch singing take at the same settings.
Load this exampleLa.Speed 72 · Pitch 64 · Mouth 128 · Throat 128Singing mode on
Same text and settings as the speech reference, with singing mode enabled.
Load this exampleHear three pitches using the same La
Keep singing mode on and change only Pitch. Compare 88, 64 and 48 using La. every time. A larger raw Pitch value makes the voiced tone lower; a smaller value makes it higher. Load an example below to put its exact text, settings and mode into the experiment.
The numbers are SAM timing parameters, not hertz or note names. These three pitches are a sketch rather than a calibrated musical scale. If you need C, D or E in a particular octave, tune the exported syllables against an instrument in your audio editor.
La.Speed 72 · Pitch 88 · Mouth 128 · Throat 128Singing mode on
Singing mode on. Pitch 88 is the lowest of these three examples; it is an engine value, not a note name.
Load this exampleLa.Speed 72 · Pitch 48 · Mouth 128 · Throat 128Singing mode on
Singing mode on. Pitch 48 is the highest of these three examples.
Load this exampleReproduce this four-beat robotic chant
The finished example repeats La. at low, middle, high and middle pitch. Each source was generated separately with singing mode on, Speed 72, Mouth 128 and Throat 128. The clips start at 0.0, 0.6, 1.2 and 1.8 seconds: four beats at 100 BPM. They are not stretched or retuned; silence fills the remainder of each beat.
Download the lower, middle and higher La samples above. Reuse the middle sample on beat four. Put them on an editing timeline at the positions below, then export a 2.4-second phrase. The downloadable finished example lets you compare your arrangement with the recipe.
Scroll the table sideways on smaller screens.
| Beat | Start time | Text / Pitch |
|---|---|---|
| 1 | 0.0 seconds | La. / 88 |
| 2 | 0.6 seconds | La. / 64 |
| 3 | 1.2 seconds | La. / 48 |
| 4 | 1.8 seconds | La. / 64 |
Low, middle, high, middle. Four separate singing-mode clips placed on a 100 BPM grid; no music, reverb or pitch correction. Download this WAV to check the arrangement.
Change the tempo without accidentally changing the pitch
The time between beats is 60 divided by BPM. At 100 BPM it is 0.6 seconds; at 120 BPM it is 0.5 seconds. Move the start of each clip to the new grid while leaving the sample itself unchanged. That changes the phrase's rhythm without resampling the syllables.
If a syllable exceeds the beat length, shorten the text or lower Speed to synthesize a faster take. Do not simply cut off the ending consonant. If you stretch a clip in an editor, use its option to preserve pitch when you want the note to stay the same. SAM's Speed control is not a BPM setting.
Replace La with words one fragment at a time
Try Robot, lights, are and glowing as four separate takes at the pitches in the table. Words with more than one syllable may need a slower musical grid than the La example. Compare each word's actual duration before placing it on a beat; singing mode does not make all words equally long.
For a sustained vowel, begin with a short familiar spelling such as La. or Oh. Repeating letters can make the reciter choose extra sounds rather than hold one note. Keep normal spelling when it remains clearer, and do exact duration work in the editor. The on-page text box accepts ordinary text, not SSML or MIDI.
Keep a reproducible set of source notes
Generate and download each accepted fragment before changing its pitch. Name the files beat-01-la-p88.wav, beat-02-la-p64.wav and beat-03-la-p48.wav. Copy Link saves the text, all four parameters and the singing-mode choice, so you can regenerate a missing note.
Keep the dry syllables as well as the final arranged phrase. Add accompaniment after the rhythm works alone, then listen for clipped endings or words masked by music. For a video project that needs 48 kHz, the WAV export page can convert ordinary SAM speech directly; convert this assembled chant in your editor after arranging it.