The TTS#3 evaluation dates have changed !!
For Chinese: from 09.02.2007 to 15.02.2007.
For the other tasks: from 16.02.2007 to 22.02.2007.
TTS evaluation data are available here.
The schedule of the whole TC-STAR Evaluation Run #3 can be seen here.
TTS Evaluation Tasks
TTS evaluations will be run in 3 languages: UK English (EN), Spanish (ES) and Chinese Mandarin (CN)..
|
Modules
|
Languages
|
| Text Processing CANCELED | EN |
| Acoustic Synthesis System | EN |
| TTS System | EN, ES, CN |
| Intra-lingual Voice Conversion (IVC) |
EN, ES, CN |
| Cross-lingual Voice Conversion (CVC) |
ES (target: EN) |
| Cross-lingual Voice Conversion based found data (fCVC) | ES (target: EN) |
Evaluation of speech synthesis Modules |
|---|
Text Processing (M1) CANCELED |
|
M1.1 Non
Standard Word Normalization
|
| M1.3 POS Tagging |
|
M1.4 Grapheme-to-Phoneme
Conversion
|
Acoustic Synthesis (M3) |
|
M3.2 MOS Tests
|
TTS System (S) |
| S1 MOS Tests |
| S2 Evaluation of intelligibility in the translation scenario |
Evaluation of Research Topics |
Evaluation of Intra-lingual Voice
Conversion (IVC)
|
|
IVC1 Comparison
of speaker
identities
|
|
IVC2 Evaluation
of overall
speech quality
|
Evaluation of Cross-lingual Voice Conversion (CVC) |
|
CVC1 Comparison
of speaker
identities
|
|
CVC2 Evaluation
of overall
speech quality
|
Evaluation of Cross-lingual Voice Conversion based on found data (fCVC) |
|
fCVC1 Comparison
of speaker
identities
|
|
fCVC2 Evaluation
of overall
speech quality
|
TTS Development Data
| Data Description | Language | Files |
| Intra-lingual Voice Conversion (IVC): Conversion directions: 75 (F) -> 76 (F) 75 (F) -> 79 (M) 80 (M) -> 76 (F) 80 (M) -> 79 (M) |
EN |
Data can be downloaded from: vc-train-english.tgz (~ 2Go) If you already have the data, you can simply use the following index: vcr_en.train |
| Intra-lingual Voice Conversion (IVC): Conversion directions: 75 (F) -> 76 (F) 75 (F) -> 79 (M) 80 (M) -> 76 (F) 80 (M) -> 79 (M) |
ES | Data can be downloaded from: Part 1: vc-train-spanish_7576.tgz (~ 1Go) Part 2: vc-train-spanish_7980.tgz (~ 1Go) If you already have the data, you can simply use the following index: vcr_es.train |
| Intra-lingual Voice Conversion (IVC): Conversion directions: 01 (F) -> 03 (F) 01 (F) -> 02 (M) 02 (M) -> 03 (F) |
CN | Data can be downloaded from: vc-train-mandarin_nokia.tgz (~2Go) Part 1 : vc-train-mandarin_03.tgz (~ 1Go) Part 2 : vc-train-mandarin_04.tgz (~ 1Go) If you already have the data, you can simply use the following index: vcr_zh.train |
| Cross-lingual Voice Conversion (CVC): - The source are the Spanish voices (speakers 75 and 80); the target are the English voices (speakers 76 and 79). - Participants cannot use the information of the target in Spanish!!! only English (except that they can find some f0 features of the target in Spanish, as the f0 mean.) - Participants can use the information of the source in English. - Additionally they can use the information of the source in Spanish (but they have to communicate it in the submission: the approach requires bilingual source speakers, i.e., CVC-bl). |
ES (target: EN) | See above. |
Cross-lingual Voice Conversion based on found data (fCVC): - The target are two english politicians: Speaker 1 (CC) and Speaker 2 (DA). Data are: Audio segments (.wav), Verbatim transcriptions (.stm), Translation in Spanish (.txt). - The source is the IBM TTS based on the TC-STAR baseline voice (spk 73, male). |
ES (target: EN) | Speaker 1 (CC): Speaker1-CC.zip For the source, you can use the following development data: * TTS samples (provided by IBM): fvc_source_tts_data.tgz
|
TTS Test data
| Data Description | Language | Files |
| Text Processing (M1.1 / M1.3 / M1.4) | ||
| Text corpus | EN | CANCELED |
| Acoustic Synthesis (M3.2) | ||
|
Acoutic Synthesis (M3.2) - English Prosodic descriptions (XML) |
EN |
9 paragraphs: tcs-tts-eval3-m3.2-epps-en.zip |
| TTS (S1 / S2) | ||
|
TTS System (S1 + S2) - English Paragraphs to synthetise Format: SSML / Unicode UTF-8 |
EN |
Task S1 (40 paragraphs): tcs-tts-eval3-S1-en.zip Task S2 (160 sentences): tcs-tts-eval3-S2-en.txt |
|
TTS System (S1 + S2) - Spanish Paragraphs to synthetise Format: SSML / Unicode UTF-8 |
ES |
Task S1 (40 paragraphs): tcs-tts-eval3-S1-es.zip Task S2 (160 sentences): tcs-tts-eval3-S2-es.txt |
|
TTS System (S1) - Chinese Paragraphs to synthetise Format: SSML / Unicode UTF-8 |
CN | Task S1 (37 paragraphs): tcs-tts-eval3-S1-cn.zip |
| Voice Conversion (IVC / CVC / fCVC) | ||
|
Intra-lingual Voice Conversion (IVC) - Chinese 3 conversion directions: 01 (F) -> 03 (F) 01 (F) -> 02 (M) 02 (M) -> 03 (F) |
CN |
Source voices: Either download data from: Voice 01: vc-evaluation-mandarin_01.tgz Voice 02: vc-evaluation-mandarin_02.tgz Or (if you already have the data) simply use the following index: vcr_zh.test |
|
Intra-lingual Voice Conversion (IVC) - English 4 conversion directions: 75 (F) -> 76 (F) 75 (F) -> 79 (M) 80 (M) -> 76 (F) 80 (M) -> 79 (M) |
EN |
Source voices: Either download data from: Voice 75: vc-evaluation-english-75.tgz Voice 80: vc-evaluation-english-80.tgz Or (if you already have the data) simply use the following index: vcr_en.test |
|
Intra-lingual Voice Conversion (IVC) - Spanish 4 conversion directions: 75 (F) -> 76 (F) 75 (F) -> 79 (M) 80 (M) -> 76 (F) 80 (M) -> 79 (M) |
ES |
Source voices: Either download data from: Voice 75: vc-evaluation-spanish-75.tgz Voice 80: vc-evaluation-spanish-80.tgz Or (if you already have the data) simply use the following index: vcr_es.test |
|
Cross-lingual Voice Conversion (CVC) - Spanish to English The sources are the Spanish voices (speakers 75 and 80) The targets are the English voices (speakers 76 and 79) |
Source: ES Target: EN |
See IVC - Spanish |
|
Cross-lingual Voice Conversion based on found data (fCVC) - Spanish to English The source is the IBM TTS based on the TC-STAR baseline voice (spk 73, male). The targets are the voices of two English politicians. |
Source: ES Target: EN |
Source voice (50 sentences): Voice 73 (IBM TTS): IBM_VC_ES_EvalSet.tar.bz2 |
TTS Participants
Available soon...
TTS Results
Available soon...