@현우 오 The concurrency limit is set at the endpoint level. Every endpoint has an individual limit so if you are using multiple endpoints and an increase is required then the limit needs to raised for each of the endpoint. You can check the note in the documentation from the referenced link in your post.
If you use custom models, please be aware, that one Speech resource may be associated with many custom endpoints hosting many custom model deployments. Each Custom endpoint has the default number of concurrent request limit (100) set by creation. If you need to adjust it, you need to make the adjustment of each custom endpoint separately. Please also note, that the value of the number of concurrent request limit for the base model of a Speech resource has no effect to the custom endpoints associated with this resource.