Update README.md
Browse files
README.md
CHANGED
@@ -19,7 +19,7 @@ grammatical errors intact was essential.
|
|
19 |
Existing speech-to-text (STT) models like Whisper tend to correct grammatical errors due to their strong internal language models, making them unsuitable for this task.
|
20 |
Therefore, SESGE was created to train a custom STT model that could accurately transcribe spoken English with grammatical errors preserved.
|
21 |
|
22 |
-
## Dataset
|
23 |
|
24 |
Given the absence of a suitable dataset for training an error-preserving STT system, DeMINT fine-tuned a Whisper model with data from two primary sources:
|
25 |
|
@@ -40,7 +40,7 @@ SESGE dataset. This means that while COREFL data was used during our training, o
|
|
40 |
|
41 |
Training samples comprise 28,592 utterances from C4_200M. Validation and test sets contain 700 samples each.
|
42 |
|
43 |
-
##
|
44 |
|
45 |
Two models were trained on the SESGE dataset by fine-tuning Whisper, enabling error-preserving STT. These models are available on the Hugging Face Hub:
|
46 |
|
@@ -50,7 +50,7 @@ Two models were trained on the SESGE dataset by fine-tuning Whisper, enabling er
|
|
50 |
Both models have been optimized to transcribe spoken English while retaining grammatical errors, making them suitable for language-learning applications
|
51 |
where fidelity to spoken errors is essential.
|
52 |
|
53 |
-
## How to
|
54 |
|
55 |
If you use the SESGE dataset, please cite the following paper:
|
56 |
|
|
|
19 |
Existing speech-to-text (STT) models like Whisper tend to correct grammatical errors due to their strong internal language models, making them unsuitable for this task.
|
20 |
Therefore, SESGE was created to train a custom STT model that could accurately transcribe spoken English with grammatical errors preserved.
|
21 |
|
22 |
+
## Dataset description
|
23 |
|
24 |
Given the absence of a suitable dataset for training an error-preserving STT system, DeMINT fine-tuned a Whisper model with data from two primary sources:
|
25 |
|
|
|
40 |
|
41 |
Training samples comprise 28,592 utterances from C4_200M. Validation and test sets contain 700 samples each.
|
42 |
|
43 |
+
## Derived models
|
44 |
|
45 |
Two models were trained on the SESGE dataset by fine-tuning Whisper, enabling error-preserving STT. These models are available on the Hugging Face Hub:
|
46 |
|
|
|
50 |
Both models have been optimized to transcribe spoken English while retaining grammatical errors, making them suitable for language-learning applications
|
51 |
where fidelity to spoken errors is essential.
|
52 |
|
53 |
+
## How to cite this work
|
54 |
|
55 |
If you use the SESGE dataset, please cite the following paper:
|
56 |
|