Tried to get bengali transcript with whisper #2054

dippreneurlab · 2024-03-02T16:49:13Z

dippreneurlab
Mar 2, 2024

Currently tried to get bengali transcript from audio, but whisper support bengali as per docs. but I got API response that, language 'bn' not supported. code -


<?php
    if ($_SERVER['REQUEST_METHOD'] === 'POST' && isset($_POST['submitRecording'])) {
        $apiKey = 'sorry dude';
        if (isset($_FILES['audioFile'])) {
            $audioFile = $_FILES['audioFile'];
            $supportedFormats = ['flac', 'm4a', 'mp3', 'mp4', 'mpeg', 'mpga', 'oga', 'ogg', 'wav', 'webm'];
            $fileExtension = pathinfo($audioFile['name'], PATHINFO_EXTENSION);
            if (in_array(strtolower($fileExtension), $supportedFormats)) {
                $apiEndpoint = 'https://api.openai.com/v1/audio/transcriptions';
                $headers = [
                    'Authorization: Bearer ' . $apiKey,
                ];
                $file = new CURLFile($audioFile['tmp_name'], mime_content_type($audioFile['tmp_name']), basename($audioFile['name']));
    
                $fileData = [
                    'file' => $file,
                    'model' => 'whisper-1',
                    'language' => 'bn',
                    'temperature' => 0.2
                ];
                $ch = curl_init($apiEndpoint);
                curl_setopt($ch, CURLOPT_POST, 1);
                curl_setopt($ch, CURLOPT_POSTFIELDS, $fileData);
                curl_setopt($ch, CURLOPT_HTTPHEADER, $headers);
                curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
                $response = curl_exec($ch);
                if (curl_errno($ch)) {
                    $result = ['error' => 'Error: ' . curl_error($ch)];
                } else {
                    $result = json_decode($response, true);
                }
                curl_close($ch);
            } else {
                $result = ['error' => 'Unsupported file format. Supported formats: ' . implode(', ', $supportedFormats)];
            }
        } else {
            $result = ['error' => 'No audio file uploaded.'];
        }
    }
?>

<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Whisper Try</title>
</head>
<body>
    <?php if (isset($result)): ?>
        <?php if (isset($result['error'])): ?>
            <p>Error: <?php echo $result['error']['message']; ?></p>
        <?php else: ?>
            <p>Text: <?php echo $result['text']; ?></p>
        <?php endif; ?>
    <?php endif; ?>
    <form method="post" enctype="multipart/form-data">
        <label for="audioFile">Choose an audio file:</label>
        <input type="file" name="audioFile" id="audioFile" accept="audio/*">
        <br>
        <button type="submit" name="submitRecording">Submit Recording</button>
    </form>
</body>
</html>

is there any problem?

glangford · 2024-03-02T20:48:14Z

glangford
Mar 2, 2024

This discussion board is the open source version of whisper, which does list bn as a supported language.

You are using the openAI remote endpoint, which is a slightly different implementation.

According to the openAI speech to text API, Bengali is not in the list of supported languages listed here
https://platform.openai.com/docs/guides/speech-to-text/supported-languages

...but then says

While the underlying model was trained on 98 languages, we only list the languages that exceeded <50% word error rate (WER) which is an industry standard benchmark for speech to text model accuracy. The model will return results for languages not listed above but the quality will be low.

So there is conflicting information here, which maybe can be resolved if you post to the OpenAI Developer Forum page. You might try not specifying any language to see if there is a change.

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Tried to get bengali transcript with whisper #2054

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{editor}}'s edit

{{editor}}'s edit

Uh oh!

Replies: 1 comment

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{editor}}'s edit

{{editor}}'s edit

Uh oh!

Select a reply

Uh oh!

Tried to get bengali transcript with whisper #2054

Uh oh!

Uh oh!

dippreneurlab Mar 2, 2024

Replies: 1 comment

Uh oh!

Uh oh!

glangford Mar 2, 2024

dippreneurlab
Mar 2, 2024

glangford
Mar 2, 2024