SR Http
- 19 Jul 2024
- 3 Minutes to read
- Contributors
- Print
- PDF
SR Http
- Updated on 19 Jul 2024
- 3 Minutes to read
- Contributors
- Print
- PDF
Article summary
Did you find this summary helpful?
Thank you for your feedback
Performs speech recognition. Uses SestekSR with HTTP protocol.
Parameters:
default | description | default |
|---|---|---|
| Address | Address of SestekSR that will be used for performing recognitions | http://core-sr |
| IgnoreSslErrors | ignore any certificate errors if SestekSR address contains https | false |
| ConnectionTimeoutMsec | SestekSR Connection timeout in terms of milliseconds | 5000 |
| CredentialId | Id of the credential that will be used for SestekSR recognitions. This parameter is only necessary if a cloud SestekSR deployment is used. You must leave this empty for on-premise deployments. | None(On-Prem) |
| Models | list of models that can be used in the recognitions | |
| Language | Language of the model that will be used for recognition | false |
| ProduceRecognizedWords | If set to true recognition events will provide extra info for each recognized word | false |
Inputs
Audio:
Accepts audio from a single channel.
Events:
name | description | known nodes that generate this event |
|---|---|---|
| Language Change | Changes the language parameter. The recognitions requested after this event will use the updated language (hence appropriate model) | Language Identifier, Text Language Identifier |
| Start Of Segment | Signals the start of a speech segment. Useful in batch mode. | Audio Segmenter |
| Speech Started | Signals the start of speech in realtime. Necessary for stream mode. Can be used in batch mode. | Vad |
| End Of Segment | Signals the start of a speech segment. Useful in batch mode. | Audio Segmenter |
| Speech Ended | Signals the end of speech in realtime. Necessary for stream mode. Can be used in batch mode. | Vad |
Outputs
Audio:
none
Events:
name | description |
|---|---|
| SR Milestone | Raised for each piece of speech in the provided audio data. This event contains recognition results for a single piece of speech. "SR Milestone" results are cumulative. The total recognition result in the end can be prepared by concatenating all the sr-milestone events' texts. |
Remarks :
- If you want to split audio and perform your recognitions you may or may not want to add an AudioSegmentor node before this node for splitting the audio into pieces.
- We recommend using end-to-end models for recognitions performed with this node.
License : Needs "sr-rest" option.
Supported flow types: Stream, Batch
Service Link Parametrization:
The default address for this node is http://core-sr . This default link is intended for on-premise installations.
When SestekSr is deployed in the cloud SestekSR uses a http address that depends on the name of the model it hosts. eg.
http://core-sr-turkish
http://core-sr-english
http://core-sr-arabic
In order to use a single link in the cloud that you won't need to change every time you switch your models you should prefer the link with the {model_title} placeholder. ie. your default link should be the following: http://core-sr-{model_title}
Project Structure
The simplest project utilizing SR Http can be built as such:
About this project
Note that the project above will not split the audio. All of the inputted audio will be sent to the SR service in bulk, and the output will be just one SR Milestone event including all of the recognized speech.
A better project that gives separate results for each speech fragment can be built like the examples below:
About Audio Splitters
One of the most complex examples utilizing the SR HTTP node can be built as such:
In the project above, every speech segment's language is identified first. Then, the speech will be recognized in the identified language. Then, the transcript will be normalized, punctuated, and filtered. At the same time, every SR Milestone event will be translated into a targeted language. All of these will be done in real-time.
At the end of the flow, all of the accumulated SR Milestone events will be used for Topic Detection, and Summarization.
Compatibility
Almost every node that utilizes textual data is compatible with the SR Http and SR WebSocket nodes' event outputs. Which is why we are able to create complex projects such as the example above utilizing these nodes.
Supported flow types: Stream, Batch
Was this article helpful?