Jürgen Trouvain

Tempo Control in Speech Synthesis by Prosodic Phrasing

Additional information

---

Text

Die SPD hat bestätigt, in der Kölner Spendenaffäre doch sämtliche Namen der Empfänger gefälschter Spendenquittungen zu kennen. Die Partei teilte in Düsseldorf und Berlin mit, die Liste sei am 10. April eingetroffen. Bislang hatte die Partei behauptet, die Namen nicht vollständig zu kennen.

---

Procedure for break/pause adjustment

Step 1

slowing down:insert minor prosodic breaks (level "2") after each syntactic noun pharse and after each syntactic adjective phrase.

speeding up: skip predicted breaks if prosodic phrase of level "3" is shorter than 6 syllables

Step 2

Pause durations (in ms) according to prosodic break strength and envisaged tempo

Break level (ToBI)	60% very fast	80% rather fast	100% default	120% rather slow	140% very slow
"2"	-	-	-	120	200
"3"	20	80	120	200	410
"4"	50	100	200	410	700
"6"	100	200	410	700	1000

Step 3

Linear time-scaling (to reach the desired total duration/speaking rate)

---

Stimuli

All audio examples in wav-format, sampling rate: 16 kHz

very fast speech (60%):

linear

adjusted breaks/pauses

---

rather fast speech (80%):

linear

adjusted breaks/pauses

---

rather slow speech (120%):

linear

adjusted breaks/pauses

"bonus tracks" rather slow speech (120%):

pauses between words

just default pauses expanded

---

very slow speech (140%):

linear

adjusted breaks/pauses

"bonus tracks" very slow speech (140%):

pauses between words

just default pauses expanded