11-752: week 8 exercises hints

Talking clock
- See http://www.festvox.org/festvox/ which has a whole chapter with a walkthrough on this task. 5.6
- Speech lab machines are set up to do this, but they are slow, though even the autolabelling, the longest part, only took me 10-15 minutes on those machines.
- Don't worry too much about recording quality, as long as its basically understandable, don't worry about background noise.
- For interest you could try delivering it in different style too.
Uniphone synthesizer, the build process is similar but a little different. See section 12.2
Set up directory structure
```
mkdir uniphone
cd uniphone
$SPPDIR/src/unitsel/setup_clunits cmu us YOURNAME uniphone
```
Prompts are in etc/uniphone.data. Build the synthesis prompts with
```
festival -b festvox/build_clunits.scm '(build_prompts "etc/uniphone.data")'
```
Record the prompts
```
bin/prompt_them etc/data 1
```
Label the utterances
```
bin/make_labs prompt-wav/*.wav
```
Hand correct the labels (in emulabe selected open and hit enter in the new window once you've selected it). Double click on the fileids
```
emulabel etc/emu_lab
```
Extract pitchmarks and Mel Cepstral Coefficients
```
bin/make_pm_wave wav/*.wav
bin/make_mcep wav/*.wav
```
Integrate the labelled durations into the utterances
```
festival -b festvox/build_clunits.scm '(build_utts "etc/uniphone.data")'
```
Build the unit selection synthesizer
```
festival -b festvox/build_clunits.scm '(build_clunits "etc/uniphone.data")'
```
Build the lpc coefficients so we can do ptich and duration modification.
```
bin/make_lpc wav/*.wav
```
Run the synthesizer voice
```
festival festvox/cmu_us_YOURNAME_clunits.scm
...
festival> (voice_cmu_us_YOURNAME_clunits)
...
festival> (SayText "Hello world")
...
```
To change the pitch range edit festvox/cmu_us_YOURNAME_f0model.scm change the target_f0_mean and stddev. Duration can be controlled in festvox/cmu_us_YOURNAME_duration.scm (the duration stretch bit).
A walkthrough of a clustergen voice build is available here