Update README.md
Browse files
README.md
CHANGED
@@ -61,7 +61,7 @@ model-index:
|
|
61 |
---
|
62 |
# Wav2Vec2-Large-Ru-Golos-With-LM
|
63 |
|
64 |
-
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Russian using the [Sberdevices Golos](https://huggingface.co/datasets/SberDevices/Golos). The language model is based on [the Russian National Corpus](https://ruscorpora.ru/
|
65 |
|
66 |
## Usage
|
67 |
|
@@ -105,7 +105,8 @@ def speech_file_to_array_fn(batch):
|
|
105 |
|
106 |
test_dataset = test_dataset.map(speech_file_to_array_fn)
|
107 |
|
108 |
-
inputs = processor(test_dataset["speech"], sampling_rate=16_000,
|
|
|
109 |
with torch.no_grad():
|
110 |
logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
|
111 |
predicted_sentences = processor.batch_decode(
|
|
|
61 |
---
|
62 |
# Wav2Vec2-Large-Ru-Golos-With-LM
|
63 |
|
64 |
+
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Russian using the [Sberdevices Golos](https://huggingface.co/datasets/SberDevices/Golos). The language model is based on [the Russian National Corpus](https://ruscorpora.ru/), and this model includes unigrams, bigrams and trigrams.
|
65 |
|
66 |
## Usage
|
67 |
|
|
|
105 |
|
106 |
test_dataset = test_dataset.map(speech_file_to_array_fn)
|
107 |
|
108 |
+
inputs = processor(test_dataset["speech"], sampling_rate=16_000,
|
109 |
+
return_tensors="pt", padding=True)
|
110 |
with torch.no_grad():
|
111 |
logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
|
112 |
predicted_sentences = processor.batch_decode(
|