We’ve prepared an audio file and a file containing two Flow Designer Flows for you to download and use with this guide. Follow the instructions listed below to load them into your own BLOCKS environment.

The first Flow is a very simple example of using the Cloud Speech-to-Text API to convert an audio file into text.

By simply placing a Speech recognition BLOCK from the Machine Learning category, we can create a text transcription of an audio file that has been uploaded to GCS.

The chart below lists the property settings for each BLOCK (only those that differ from the default settings). Each BLOCK also contains an explanation for what it does in its “BLOCK memos” property, but these are not included in this chart.

Click the play_circle_outline button within the Start of Flow BLOCK's (BLOCK name: Cloud Speech API sample (1)) properties to execute the Flow.

If successful, a log similar to the one shown below will be output to the Logs section.

{
  "results": [
    {
      "alternatives": [
        {
          "transcript": "with an intuitive design and affordable pricing plans blocks makes getting started with machine learning simple now businesses can make data-driven decisions themselves without the need for expensive services in expert help",
          "confidence": 0.9624174
        }
      ]
    }
  ],
  "gcs_url": "gs://my-bucket/speech_api_sample_voice_en.flac",
  "timestamp": 1497943130.0
}
 

"transcript": "with an intuitive design...in expert help" This portion contains the results of transcribing the audio file.

"confidence": 0.9624174 This portion contains a value showing the level of confidence that the results are correct. The confidence level is on a scale of 0.0 to 1.0 with a higher number showing a higher confidence. Generally, this is only output for the text with the highest confidence level.

These results have about a 96% confidence level, and there was a bit that wasn't transcribed correctly (shown in red letters above).

By entering important vocabulary or phrases into the Contextual word/phrase hints property of the Speech recognition BLOCK, we can raise the accuracy level of the transcription.

We'll be doing this in the second example Flow, as well as storing the transcription results into BigQuery.

The chart below lists the property settings for each BLOCK (only those that differ from the default settings). Each BLOCK also contains an explanation for what it does in its “BLOCK memos” property, but these are not included in this chart.

[
 {
  "name": "results",
  "type": "RECORD",
  "mode": "REPEATED",
  "fields": [
   {
    "name": "alternatives",
    "type": "RECORD",
    "mode": "REPEATED",
    "fields": [
     {
      "name": "transcript",
      "type": "STRING",
      "mode": "NULLABLE"
     },
     {
      "name": "confidence",
      "type": "FLOAT",
      "mode": "NULLABLE"
     }
    ]
   }
  ]
 },
 {
  "name": "gcs_url",
  "type": "STRING",
  "mode": "NULLABLE"
 },
 {
  "name": "timestamp",
  "type": "TIMESTAMP",
  "mode": "NULLABLE"
 }
]

SELECT
  results.alternatives.transcript as transcript
FROM
  [blocks_samples.speech_api]
WHERE
  results.alternatives.confidence > 0
ORDER BY
  timestamp desc
LIMIT
  1

Click the play_circle_outline button within the Start of Flow BLOCK's (BLOCK name: Cloud Speech API sample (2)) properties to execute the Flow.

If successful, a log similar to the one shown below will be output to the Logs section.

[
  {
    "transcript": "with an intuitive design and affordable pricing plans blocks makes getting started with machine learning simple now businesses can make data-driven decisions themselves without the need for expensive services and expert help"
  }
]

As with the previous example, the "transcript": "with an intuitive...and expert help" portion contains the results of transcribing the audio file. The results have changed from the first sample since we entered values into the Speech recognition BLOCK's Contextual word/phrase hints property (shown in blue letters above).

We'll briefly explain in this section how we recorded the sample audio file. You can use this for reference to test things with your own audio data.

There are built-in applications in Windows and macOS (OS X) that you can use to record audio. However, since these do not record in a format supported by the Speech recognition BLOCK, you cannot use these files right away. You will need to install a separate application to convert audio files into a format supported by the Speech recognition BLOCK.

We will introduce how to install the program Audacity open_in_new and use it to record audio. Audacity is an open-source program for recording and editing audio that can output files in the FLAC format supported by the Speech recognition BLOCK.

Instructions for installing and using Audacity to record data will be explained in the following order:

Recording audio with Audacity

The Audacity screen is divided into various sections. The sections required for recording audio and their names are shown in the image below.

info_outline For macOS/OS X users, the menu bar will be at the top of the screen, rather than within the Audacity window.

The process for recording audio is as follows:

1. Set the “Recording Device” and “Recording Channel” from the Device Toolbar.

   

   • Under “Recording Device”, choose the device you will use to make the recording.
   • Under “Recording Channel”, select 1 (Mono) Recording Channel.

2. From the Selection Toolbar in the Lower Toolbar dock area, set the Project Rate (Hz) to 16000 as recommended for the Cloud Speech API.

   

3. With that, you are ready to record the audio. However, we will first check the audio levels (without recording) by clicking the Recording Meter Toolbar.

   

4. Adjust the recording level using the Recording Slider portion of the Mixer Toolbar.

   

5. After adjusting the recording level, click the Recording Meter Toolbar again to check the recording level. Repeat these steps as needed.

6. Click the Record button from the Transport Toolbar to begin recording. Press the Stop button to stop the recording.

   

7. To export the audio as a FLAC format file, open the File menu and select Export Audio.

   

8. From the Export Audio window, set each required item and click the Save button.

   

   Refer to the following list for each item (macOS/OS X terms shown in parentheses):

   • File name (Save as): Set the file name.
   • Save in (Where): Set the location where the file will be saved.
   • Save as type (File type): Select FLAC Files.
   • Level: Set the compression level between 0 and 8. The default level is 5. 0 refers to no compression, and the compression rate increases with each higher number.
   • Bit depth: Select 16 bit.