Towards Cross-Lingual Voice Adaptation for Conversational Speech
MS Thesis

Read Speech TTS versus Conversational Speech TTS

Examples	Systems
	Read Speech TTS	Conversational TTS
Example 1
Example 2
Example 3
Example 4
Example 5

Read Speech TTS is trained using read speech data. Conversational TTS is trained using classroom lecture data.

Voice Adaptation Experiments

Preliminary Experiment 1 : Source Content Disentanglement

Examples	Systems
	Source	Target	Generated
Example 1

Preliminary Experiment 2 - Adaptation : Read Speech VC versus Conversational Speech VC

Examples	Systems
	Original	Adapatation data 7mins
Read Speech VC
Conversational Speech VC

Experiment 1 : Manually Curated Data

Monolingual

Examples	Systems
	Original	E2E	HTS
Example 1
Example 2
Example 3
Example 4
Example 5

Cross-Lingual

Examples	Systems
	Original	E2E	HTS	HTS + Cycle_Gan
Example 1
Example 2
Example 3
Example 4
Example 5

Generated audios are the hindi translations of the target audios
We are adapting from Bilingual (Hindi+English) speaker to the target speaker

Experiment 2 - Pruning

Speaker 1

Examples	System Comparison
	Original	English	Hindi	Kannada
Example 1
Example 2
Example 3
Example 4
Example 5

Speaker 2

Examples	Systems
	Original	English	Hindi	Kannada
Example 1
Example 2
Example 3
Example 4
Example 5

Speaker 3

Examples	Systems
	Original	English
Example 1
Example 2
Example 3
Example 4
Example 5

Videos better when opened in firefox browser