rossenbach-add-lm-info-binaries (#3)

- add language resourced readme section (38fcb15b2ea5022bce494556de8261fd5493f2a3)
- Upload compressed ARPA LMs (c97aceb445a63dd16a1963c865a54867ef1cc0ce)
- Add binary ARPA langauge model files (293b9bfb83ac6f7005cbcb934d42e561420e2a40)


Co-authored-by: Nick Rossenbach <[email protected]>

3gram-pruned.arpa.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa6ca8547df5d2577ddee405501e8cf51a579429c6c6fd159a91ab81fd039a7a
+size 755342620

3gram-pruned.arpa.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a1e697b1c3afeb6660434173f929104d7456ae62b065f9f13f309ecd408dd2f3
-size 1022207015
+oid sha256:ee3f7c8c013f5ba019f988860187ed0b94670e09cdc24b6e3e98ab25f298829d
+size 346396941

4gram-pruned.arpa.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b61b961ba0238737fd0ca74d3561b720adda31e5666c182001a490f4e4f46612
+size 1204786332

4gram-pruned.arpa.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f968a6c108d19d966afa562b08deac2f74167f393bb1bd56609ca4835ae140dc
-size 1737364471
+oid sha256:f5d83ff14540b3c2486741d3e3d8fe0305eb34ca88fb675ec243a69bc513787f
+size 563213596

4gram-unpruned.arpa.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db7836e28471f4d944ac84f92352ad8e0bf136e17f0dfcc262a2ad59f192eaea
+size 5022244796

4gram-unpruned.arpa.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd511ce814b9fc2d00dc709de8ce2e2d9d998bfb5326c1dbb203e6546acd3bc7
-size 8966175527
+oid sha256:e64b2328541f7bf7beeebffccf0d2075b1f0a8fed4b291c53550b2972838ae1b
+size 2511206942

README.md

@@ -185,6 +185,21 @@ Some of the above datasets, in particular People's Speech, Yodas and CommonVoice
 
 Audios are embedded as raw bytes (can be decoded by soundfile). We chunked and created smaller audio files from long ones based on start and stop supervision from the different manifests of the datasets (this is necessary for HuggingFace). Language ID with a [SpeechBrain language ID model](https://huggingface.co/speechbrain/lang-id-voxlingua107-ecapa) was performed on Yodas.
 
+### Count-based Language Models and Lexicon
+
+The dataset includes three ARPA-format count-based language models trained on the text of the *train.large* subset of Loquacious:
+
+| LM  | Size | Binary Size | Dev. Perplexity |
+| --- | --- | --- | --- |
+| 3-gram pruned |   331MB | 721MB | 222 |
+| 4-gram pruned |   538MB | 1.2GB | 202 |
+| 4-gram unpruned | 2.4GB | 4.7GB | 193 |
+
+Each language model is limited to a vocabulary containing 216k words.
+
+We also provide a pronunciation lexicon using ARPA-style phonemes containing one or multiple pronunciations for each of the words in the vocabulary.
+The original pronunciations are based on [CMUDict 0.7b](http://www.speech.cs.cmu.edu/cgi-bin/cmudict). Missing pronunciations were generated using [Sequitur](https://github.com/sequitur-g2p/sequitur-g2p), for which we also provide the trained G2P model.
+
 #### Referencing the Loquacious Set and SpeechBrain
 
 ```