Azure speech Translation not doing translation properly

Hello Miss Nansy,

Welcome to Microsoft Q&A,

Speech Translation returns both the source transcript and the translations. In your handler, read result.Translations[<lang>] (or e.result.translations['fr'] in JS), not just result.Text.

For a longer speech, Use StartContinuousRecognitionAsync() (or startContinuousRecognitionAsync) instead of a single RecognizeOnce* call. Single-shot ends after one utterance (~silence or ~30 s).

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-recognize-speech?pivots=programming-language-csharp

Increase the segmentation silence timeout so short pauses don’t end the phrase too early (e.g., 1200–2000 ms): speechConfig.SetProperty(PropertyId.Speech_SegmentationSilenceTimeoutMs, "1500"); Valid range: 100–5000 ms.

https://learn.microsoft.com/en-us/python/api/azure-cognitiveservices-speech/azure.cognitiveservices.speech.translation.translationrecognizer?view=azure-python

Please upvote and accept the answer if it helps!!

Hi Miss,

I've tested the speech translation service deployed in eastus2 on my end; it is working fine. You can use the below code to test it at your end:

import azure.cognitiveservices.speech as speechsdk
import threading
import time

speech_key = "<YOUR-SPEECH-KEY>"
service_region = "<DEPLOYMENT-REGION>"

translation_config = speechsdk.translation.SpeechTranslationConfig(
    subscription=speech_key,
    region=service_region
)
translation_config.speech_recognition_language = "en-US"
translation_config.add_target_language("es")  # Spanish

audio_config = speechsdk.audio.AudioConfig(use_default_microphone=True)

translator = speechsdk.translation.TranslationRecognizer(
    translation_config=translation_config,
    audio_config=audio_config
)

last_speech_time = time.time()
silence_timeout = 5  # seconds
stop_flag = False

output_log = []

def recognizing_handler(evt):
    global last_speech_time
    if evt.result.text.strip():
        last_speech_time = time.time()
        print(f"Recognizing: {evt.result.text}", end="\r")

def recognized_handler(evt):
    global last_speech_time
    text = evt.result.text.strip()
    if text:
        last_speech_time = time.time()
        print(f"\nRecognized: {text}")
        entry = f"Input: {text}\n"
        for lang, translation in evt.result.translations.items():
            translated_text = f"Output ({lang}): {translation}"
            print(translated_text)
            entry += translated_text + "\n"
        output_log.append(entry)

def silence_watcher():
    global stop_flag
    while not stop_flag:
        time.sleep(1)
        if time.time() - last_speech_time > silence_timeout:
            print(f"\nNo speech detected for {silence_timeout}s — stopping translation.")
            translator.stop_continuous_recognition()
            stop_flag = True
            break

translator.recognizing.connect(recognizing_handler)
translator.recognized.connect(recognized_handler)

print("Speak into your microphone...")
print("(It will keep listening and stop 5 seconds after your last word.)")

translator.start_continuous_recognition()

watcher_thread = threading.Thread(target=silence_watcher)
watcher_thread.daemon = True
watcher_thread.start()

while not stop_flag:
    time.sleep(0.1)

print("Translation session ended.")

with open("translation_output.txt", "w", encoding="utf-8") as f:
    for entry in output_log:
        f.write(entry)
        f.write("\n" + "-"*40 + "\n")

print("Output saved to translation_output.txt")

The above code will keep listening continuously and will stop after 5 seconds of silence.

Here is the supported documentation:
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-translate-speech?tabs=terminal&pivots=programming-language-python

You can also check Audio Seconds translated on the Azure Portal, as shown below:

Feel free to accept this as answer.

Thankyou for reaching out to the Microsoft QNA Portal.