How to Use Speech to Text Converter

Vovsoft Speech to Text Converter supports offline and online speech engines.

• 💻Offline
  1. Vosk
  2. Continuous Dictation
• 🌎Online
  1. OpenAI (Whisper)
  2. Deepgram
  3. Microsoft Azure
  4. IBM Cloud

💻 Offline Speech Engines

Offline speech recognition engines do not require internet connection and they run locally on your computer. However, they tend to be less accurate than online models, especially with complex speech or accents. Additionally, they can consume more local resources, potentially slowing down your system.

1. Vosk

How to use Vosk (Offline - File to Text)

Vosk is a speech recognition toolkit that works offline, supporting 20+ languages. It performs speech-to-text conversion on your own computer. No file is sent to internet in any case.

Vosk requires a "models" directory, which includes language data. Vovsoft Speech to Text Converter embeds lightweight English and French models. If you need other languages or not satisfied with the results, please follow the steps:

1. Go to https://alphacephei.com/vosk/models and download any model file for free.
2. Extract the model file into the models folder.
   Example: C:\Program Files (x86)\VOVSOFT\Speech to Text Converter\vosk\models
3. Restart Vovsoft Speech to Text Converter.

👉 Please note that .NET Framework 4.8 is required to use this feature. This version of the .NET Framework comes preinstalled on most Windows 10 and Windows 11 systems.

2. Continuous Dictation

How to use Continuous Dictation (Offline - Microphone to Text)

Continuous Dictation requires the "Microsoft Speech Platform", which is preinstalled on most systems. This feature supports English, French, German, Japanese, Simplified Chinese, Spanish, and Traditional Chinese.

How to change speech recoginiton settings:

1. Press Windows key.
2. Type and enter Control Panel.
3. Find and click Speech Recognition.

If no speech recognizer is installed on your system, please follow the steps:

1. Download and install Microsoft Speech Platform: https://www.microsoft.com/en-us/download/details.aspx?id=27225
2. Download and install Speech Recognition (SR) languages: https://www.microsoft.com/en-us/download/details.aspx?id=27224
3. Connect microphone (if not already connected).
4. Restart computer.

🌎 Online Speech to Text API Services

If you want to perform speech-to-text conversion on cloud servers instead of your own computer and take advantage of the latest AI advancements, you will need credentials from at least one of these providers:

1. OpenAI (Whisper)

How to get OpenAI API Key

In order to get your OpenAI API key, please follow these steps:

1. Go to https://platform.openai.com/signup and create your OpenAI account for free.
2. Go to https://platform.openai.com/account/api-keys and create your API key.

👉 What is the maximum file size?
OpenAI Whisper's file size limit is 26,214,400 bytes (25MB).

👉 What is API Temperature?
The sampling temperature, between 0 and 1. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.

2. Deepgram

How to get Deepgram API Key

In order to get your Deepgram API key, please follow these steps:

1. Go to https://console.deepgram.com/signup and create your Deepgram account for free.
2. Click API Keys.
3. Click Create a New API Key.

3. Microsoft Azure

How to get Microsoft Azure API Key and API Region

In order to get your Microsoft Azure API key and API Region, please follow these steps:

1. Go to https://portal.azure.com and create your Microsoft Azure account for free.
2. Click Create a resource.
3. Choose Cognitive Services.
4. After you created your Speech Service, your credentials can be found in Keys and Endpoint page: KEY1, KEY2 (any one of the keys should work) and Location/Region fields.

Credentials screen of Microsoft Azure

Supported Languages in Microsoft Azure

Afrikaans, Albanian, Amharic, Arabic, Armenian, Azerbaijani, Bengali, Bosnian, Bulgarian, Burmese, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Filipino, Finnish, French, Georgian, German, Greek, Gujarati, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Irish, Italian, Japanese, Javanese, Kannada, Kazakh, Khmer, Korean, Lao, Latvian, Lithuanian, Macedonian, Malay, Malayalam, Maltese, Marathi, Mongolian, Nepali, Norwegian Bokmål, Pashto, Persian, Polish, Portuguese, Romanian, Russian, Serbian, Sinhala, Slovak, Slovenian, Somali, Spanish, Swahili, Swedish, Tamil, Telugu, Thai, Turkish, Ukrainian, Uzbek, Vietnamese, Welsh, and Zulu are supported in Microsoft Azure Cognitive Services.

4. IBM Cloud

How to get IBM Cloud API Key and API URL

In order to get your IBM Cloud API key and API URL, please follow these steps:

1. Go to https://cloud.ibm.com/registration and create your IBM Cloud account for free.
   Please Note: If you get this error message "Your account cannot be created at this time", use your Gmail email address. It seems like IBM Watson doesn't like some email providers. So, if you have any such problems, just use another email address.
    
2. Go to https://cloud.ibm.com/catalog/services/speech-to-text and create your Speech to Text Lite Plan instance.
    
3. Go to https://cloud.ibm.com/resources; under AI / Machine Learning tab, click on your Speech to Text instance. Your credentials (API key and URL) will be displayed in Manage or Service credentials page.

Credentials screen of IBM Cloud

Enter your API key and URL into the Settings panel inside "Vovsoft Speech to Text Converter". The software is now ready to convert audio to text.

Supported Languages and Models in IBM Cloud

English, Arabic, Chinese (Mandarin), Czech, Dutch, French, German, Hindi (Indian), Italian, Japanese, Korean, Portuguese (Brazilian) and Spanish are supported in IBM Cloud API.

For most languages, the IBM Cloud service supports broadband, narrowband, telephony and multimedia models:

• Broadband models are for audio that is sampled at greater than or equal to 16 kHz.
• Narrowband models are for audio that is sampled at 8 kHz. Use narrowband models for offline decoding of telephone speech, which is the typical use for this sampling rate.
• Telephony models are intended specifically for audio that is communicated over a telephone. Like previous-generation narrowband models, telephony models are intended for audio that has a minimum sampling rate of 8 kHz.
• Multimedia models are intended for audio that is extracted from sources with a higher sampling rate, such as video. Use a multimedia model for any audio other than telephonic audio. Like previous-generation broadband models, multimedia models are intended for audio that has a minimum sampling rate of 16 kHz.

Choosing the correct model is important. Use the model that matches the sampling rate (and language) of your audio. The service automatically adjusts the sampling rate of your audio to match the model that you specify. More information: https://cloud.ibm.com/apidocs/speech-to-text

Approximate Conversion Time

Conversion times are listed in the table below. Please note that the specified times vary depending on the content of the file, its quality, language model, load of the AI servers and your computer's upload speed.

Common Errors

HTTP/1.1 503 Service Unavailable
Your URL is wrong. Please enter the exact "API Key" and "API URL" that was supplied for you by IBM Cloud.

Error sending data: (12030) The connection with the server was terminated abnormally
A firewall, proxy or antivirus interferes with the connection. Please try to disable them or use different internet connection.

Error reading data: (12152) The server returned an invalid or unrecognized response
Your audio is too long. Please try to convert a shorter audio.

"Please wait" hangs, nothing happens
The file size of your audio is too large. Please try to upload a smaller file. Converting stereo to mono may help.