Update README.md

README.md

@@ -90,13 +90,32 @@ The audio and transcriptions are in English, as per the TED talks at http://www.
 
 ### Data Splits
 
-The data is split into a training, validation and test set.
-
-| Split      | Examples |
-|------------|----------|
-| Train      |   56,803 |
-| Validation |      591 |
-| Test       |    1,469 |
+Release 1 (default config):
+- The first release of the TED-LIUM corpus, consisting of English-language TED talks with accompanying transcriptions, sampled at 16kHz.
+- It contains about 118 hours of speech.
+- Homepage: https://www.openslr.org/7/
+
+Release 2:
+- 1495 audio talks and their automatically aligned transcriptions.
+- Dictionary with pronunciations (159848 entries).
+- Selected monolingual data for language modeling from WMT12 publicly available corpora.
+- Homepage: https://www.openslr.org/19/
+
+Release 3:
+
+- 2351 audio talks and their automatically aligned transcriptions, totalling 452 hours of audio.
+- TED-LIUM 2 validation and test data: 19 TED talks with their corresponding manual transcriptions.
+- Dictionary with pronunciations (159848 entries), the same file as the one included in TED-LIUM 2.
+- Selected monolingual data for language modeling from WMT12 publicly available corpora: these files come from the TED-LIUM 2 release, but have been modified to produce a tokenization more relevant for English language.
+- Homepage: https://www.openslr.org/51/
+
+Each release is split into a training, validation and test set:
+
+| Split      | Release 1 | Release 2 | Release 3 |
+|------------|-----------|-----------|-----------|
+| Train      | 56,803    | 92,973    | 268,263   |
+| Validation | 591       | 591       | 591       |
+| Test       | 1,469     | 1,469     | 1,469     |
 
 
 ## Dataset Creation