Datasets:
LIUM
/

sanchit-gandhi HF staff commited on
Commit
9206b63
1 Parent(s): 18ab3aa

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +16 -0
README.md CHANGED
@@ -57,6 +57,22 @@ task_ids:
57
 
58
  The TED-LIUM corpus is English-language TED talks, with transcriptions, sampled at 16kHz. It contains about 118 hours of speech.
59
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
  ### Supported Tasks and Leaderboards
61
 
62
  - `automatic-speech-recognition`: The dataset can be used to train a model for Automatic Speech Recognition (ASR). The model is presented with an audio file and asked to transcribe the audio file to written text. The most common evaluation metric is the word error rate (WER). The task has an active leaderboard which can be found at https://paperswithcode.com/sota/speech-recognition-on-tedlium that ranks models based on their WER.
 
57
 
58
  The TED-LIUM corpus is English-language TED talks, with transcriptions, sampled at 16kHz. It contains about 118 hours of speech.
59
 
60
+
61
+ ### Example
62
+
63
+ ```
64
+ from datasets import load_dataset
65
+
66
+ tedlium = load_dataset("LIUM/minds14", "release1") # for Release 1
67
+
68
+ # see structure
69
+ print(tedlium)
70
+
71
+ # load audio sample on the fly
72
+ audio_input = tedlium["train"][0]["audio"] # first decoded audio sample
73
+ transcription = tedlium["train"][0]["text"] # first transcription
74
+ ```
75
+
76
  ### Supported Tasks and Leaderboards
77
 
78
  - `automatic-speech-recognition`: The dataset can be used to train a model for Automatic Speech Recognition (ASR). The model is presented with an audio file and asked to transcribe the audio file to written text. The most common evaluation metric is the word error rate (WER). The task has an active leaderboard which can be found at https://paperswithcode.com/sota/speech-recognition-on-tedlium that ranks models based on their WER.