Google-cloud-php: Base64-encoded files from PHP not accepted

[LoekvanKooten]

As your standard PHP script is not working at all @https://github.com/googleapis/google-cloud-php/issues/3254, I tried a cURL request instead:

This file works from the command prompt:

D:\Temp>gcloud ml speech recognize subs.wav --language-code=ja-JP
{
  "results": [
    {
      "alternatives": [
        {
          "confidence": 0.9474271,
          "transcript": "父には愛人がいたんですこのカードのことが頭から離れないんです僕はただ真実が知りたいだけです私じゃないわ"
        }
      ]
    },
    {
      "alternatives": [
        {
          "confidence": 0.90079355,
          "transcript": "男の人ちょっと怪しくないだって5000万超えた途端近づいてきたんでしょう私言ったよね子供はまだ飯行ってこれだった"
        }
      ]
    },
    {
      "alternatives": [
        {
          "confidence": 0.8782166,
          "transcript": "子供欲しいんだよあなたは誰かを好きなとことある"
        }
      ]
    }
  ]
}

However, in PHP this happens:

$file = file_get_contents('D:\Temp\subs.wav');
$audio = base64_encode($file);
$apikey = 'DELETEDONPURPOSE';

$ch = curl_init();

curl_setopt($ch, CURLOPT_URL, 'https://speech.googleapis.com/v1/speech:recognize?key=' . $apikey);
curl_setopt($ch, CURLOPT_RETURNTRANSFER, 1);
curl_setopt($ch, CURLOPT_POST, 1);
curl_setopt($ch, CURLOPT_POSTFIELDS, "{\n  'config': {\n    'encoding': 'LINEAR16',\n    'sampleRateHertz': 16000,\n    'languageCode': 'ja-JP',\n    'enableWordTimeOffsets': false\n  },\n  'audio': {\n    'content': ' . $audio . '\n  }\n}");

$headers = array();
$headers[] = 'Content-Type: application/json; charset=utf-8';
curl_setopt($ch, CURLOPT_HTTPHEADER, $headers);

$result = curl_exec($ch);
if (curl_errno($ch)) {
    echo 'Error:' . curl_error($ch);
}
curl_close($ch);

var_dump($result);

This gives:
string(4420931) "{ "error": { "code": 400, "message": "Invalid value at 'audio.content' (TYPE_BYTES), Base64 decoding failed for \" . UklGRjhLGQBX...

So all options are blocked. I have to add that compared to Azure, this is about the most user-unfirendly documentation I have ever seen. Nothing works out of the box. I got Azure Speech Regocnition up and running in 5 minutes; I've spent 2 days on Google, and nothing works. The only reason why I'm persisting, is that your voice recognition seems to be superior. But please, help me out here.

[jdpedrie]

Your request body contains invalid JSON. JSON fields and string values must use double quotes.

Use json_encode to properly encode your request body:

$body = [
    'config' => [
        'encoding' => 'LINEAR16',
        'sampleRateHertz' => 16000,
        'languageCode' => 'ja-JP',
        'enableWordTimeOffsets' => false,
    ],
    'audio' => [
        'content' => base64_encode($audio)
    ]
];
curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($body));

[LoekvanKooten]

Thank you for your help! This was generated by an automated cuRL generator on the net and I never looked back. Stupid of me.

This helps already, but we're not there yet...

Transcribing D:\Temp\subs_1min.wav...

Result:
string(3) "{} "

I double-checked: the file has a bit-depth of 16 bits, is mono and has a 16000 sampling rate. It's size is about 180 K.

Of course, this is purely for testing purposes. If it works, I'm genuinely interested in taking a Google Cloud subscription.

We currently have:

$audio = "D:\Temp\subs_8s.wav";
echo 'Transcribing ' . $audio . '...<br><br>';

$ch = curl_init();

curl_setopt($ch, CURLOPT_URL, 'https://speech.googleapis.com/v1/speech:recognize?key=' . $apikey);
curl_setopt($ch, CURLOPT_RETURNTRANSFER, 1);
curl_setopt($ch, CURLOPT_POST, 1);
$body = [
    'config' => [
        'encoding' => 'LINEAR16',
        'sampleRateHertz' => 16000,
        'languageCode' => 'ja-JP',
        'enableWordTimeOffsets' => false,
    ],
    'audio' => [
        'content' => base64_encode($audio)
    ]
];
curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($body));

$headers = array();
$headers[] = 'Content-Type: application/json; charset=utf-8';
curl_setopt($ch, CURLOPT_HTTPHEADER, $headers);

$result = curl_exec($ch);
if (curl_errno($ch)) {
    echo 'Error:' . curl_error($ch);
}
curl_close($ch);

var_dump($result);

[jdpedrie]

Are you able to share the source audio file? It's much easier to debug speech when both of us are working off the same inputs.

[LoekvanKooten]

Sure! https://www.akebono.nl/subs_8s.wav

[jdpedrie]

Your sample doesn't fetch the audio file contents -- is that the issue? You're sending the file name as the audio content. When I send the base64-encoded contents to the API, I receive a response:

string(271) "{
  "results": [
    {
      "alternatives": [
        {
          "transcript": "手には愛人がいたんですこのカードのことが頭から離れないんです",
          "confidence": 0.94493204
        }
      ],
      "languageCode": "ja-jp"
    }
  ]
}"

[LoekvanKooten]

But of course! During all the edits, that bit got lost. Thank you very much for your help! I now have a working concept. I'll continue with some larger files and then subscribe. It's strange that @#3254 doesn't work, but this does the same in the end and is good enough for me. Besides, I like this approach more, as I can now exacty see what is happening, whereas using the Google Libraries is pretty convoluted and not very transparant.

Final result:

`$audiofilename = "D:\Temp\subs_8s.wav";
$audiofilecontents = file_get_contents($audiofilename);
echo 'Transcribing ' . $audiofilename . '...

';

$ch = curl_init();

curl_setopt($ch, CURLOPT_URL, 'https://speech.googleapis.com/v1/speech:recognize?key=' . $apikey);
curl_setopt($ch, CURLOPT_RETURNTRANSFER, 1);
curl_setopt($ch, CURLOPT_POST, 1);
$body = [
'config' => [
'encoding' => 'LINEAR16',
'sampleRateHertz' => 16000,
'languageCode' => 'ja-JP',
'enableWordTimeOffsets' => false,
],
'audio' => [
'content' => base64_encode($audiofilecontents)
]
];
curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($body));

$headers = array();
$headers[] = 'Content-Type: application/json; charset=utf-8';
curl_setopt($ch, CURLOPT_HTTPHEADER, $headers);

$result = curl_exec($ch);
if (curl_errno($ch)) {
echo 'Error:' . curl_error($ch);
}
curl_close($ch);

var_dump($result);`