IssueTracker

Problem you have encountered:

Before migrating to the Conformer model[1], we can get "Katakana"(カタカナ) notation as well as "Kanji" notation (Chinese Character notation) when recognizing a person's name.

[1] https://cloud.google.com/speech-to-text/docs/conformer-migration#opt_out_and_migrate_late

Reproduction

Assumption

• The service is migrated to the Conformer Model.

Config & Command

## request.json
{
 "config": {
     "languageCode": "ja-JP",
     "encoding": "WEBM_OPUS",
     "sampleRateHertz":48000,
     "enableWordTimeOffsets": true,
     "enableAutomaticPunctuation":true,
     "model": "phone_call",
 },
 "audio": {
   "uri": "gs://path/to/my-audio-fils/sample.webm"
 }
}

## curl command
curl -X POST \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "x-goog-user-project: $PROJECT_ID" \
   -H "Content-Type: application/json; charset=utf-8" \
   -d @request.json \
   "https://speech.googleapis.com/v1/speech:recognize"

Observed output:

Item "words" only contains Chinese character notations.
(ex. ”山田", ”太郎")

{
 "results": [
   {
     "alternatives": [
       {
         "transcript": "山田太郎",
         "confidence": 0.9282486,
         "words": [
           {
             "startTime": "0s",
             "endTime": "2.500s",
             "word": "山田"
           },
           {
             "startTime": "2.500s",
             "endTime": "2.600s",
             "word": "太郎"
           },
	
    <<<....skipped....>>>
}

What you expected to happen:

Please fix/implement the service so that it returns the response with "Katakana" notaion as the old model did.

The old model return the response like this:
Item "words" contains Chinese character and Katakana notations.
(ex. ”山田 | ヤマダ", ”太郎 | タロー")

{
  "results": [ {
    "alternatives": [ {
      "confidence": 0.92143393,
      "transcript": "山田太郎",
      "words": [ {
        "confidence": 0.92383033,
        "endTime": "2.100s",
        "startTime": "0.200s",
        "word": "山田|ヤマダ"
      }, {
        "confidence": 0.91903746,
        "endTime": "2.400s",
        "startTime": "2.100s",
        "word": "太郎|タロー"
      } ]

   <<<....skipped....>>>
}

Temporal Workaround

A user can opt-out the Conformer model and use the previous model at least June, 2024.

https://cloud.google.com/speech-to-text/docs/conformer-migration?#opt_out_and_migrate_later