Commit
•
9206b63
1
Parent(s):
18ab3aa
Update README.md
Browse files
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.
|