AI APIs — Speech-to-Text & Language Translation
Modern Artificial Intelligence services are delivered over the internet via AI Cloud APIs. In this chapter, we build an end-to-end multi-cloud AI pipeline using Python: transcribing binary audio files into text with Speech to Text API (via POST requests), and translating that transcript into foreign languages with Language Translator API.
19.1 API Keys and Endpoints — Cloud Security
To use cloud AI models, services require two critical pieces of information: an Endpoint URL and an API Key.
The web location of the AI service instance on the internet.
A unique, secret string that authorizes your application and meters your usage.
Authorization: Bearer <key>). The
server validates your identity before processing.
Test how cloud servers respond when valid vs invalid API keys are transmitted in HTTP request headers.
19.2 Speech-to-Text — Transcribing Audio with Python
To convert spoken audio into written text, your program sends a POST request carrying the raw binary audio bytes to the Speech-to-Text API endpoint. The AI model analyzes the frequency spectrum and returns a structured JSON transcription.
from ibm_watson import SpeechToTextV1
from ibm_cloud_sdk_core.authenticators import IAMAuthenticator
# 1. Authenticate using API Key
authenticator = IAMAuthenticator("YOUR_API_KEY")
s2t = SpeechToTextV1(authenticator=authenticator)
# 2. Set the Service Endpoint URL
s2t.set_service_url("URL_s2t")
# 3. Open audio file in binary read mode ('rb')
filename = "hello_world.wav"
with open(filename, "rb") as audio_file:
# 4. Send POST request with binary audio payload
response = s2t.recognize(
audio=audio_file,
content_type="audio/wav"
)
# 5. Extract text from nested JSON response
results = response.get_result()
recognized_text = results["results"][0]["alternatives"][0]["transcript"]
print("Recognized Text:", recognized_text)Why mode='rb'? Audio files (like .wav or
.mp3) contain raw binary bytes, not plain text lines. Using open(filename, 'rb')
reads the exact byte stream required by the POST body.
results['results'][0]['alternatives'][0]['transcript']. The alternatives list
contains hypothesis text sorted by confidence score.
Select an audio clip, click Transcribe Audio, and watch the POST request transmit the binary audio to the Watson AI model.
19.3 Language Translator — AI Text Translation
Once you have transcribed text, you can chain it directly into the Language Translator API to
convert it into target languages such as Spanish (es), French (fr), or German
(de).
from ibm_watson import LanguageTranslatorV3
from ibm_cloud_sdk_core.authenticators import IAMAuthenticator
# 1. Authenticate Language Translator
authenticator = IAMAuthenticator("YOUR_API_KEY")
lt = LanguageTranslatorV3(
version="2018-05-01",
authenticator=authenticator
)
lt.set_service_url("URL_lt")
# 2. Translate English text to Spanish ('en-es')
translation_response = lt.translate(
text=recognized_text,
model_id="en-es"
)
# 3. Extract translated string
result = translation_response.get_result()
spanish_text = result["translations"][0]["translation"]
print("Spanish Translation:", spanish_text)
# 4. Reverse translate back to English ('es-en')
back_to_english = lt.translate(
text=spanish_text,
model_id="es-en"
).get_result()["translations"][0]["translation"]
print("Verified English:", back_to_english)Model IDs: The model_id parameter specifies
source language and target language joined by a hyphen (e.g. 'en-es' for English → Spanish,
'en-fr' for English → French).
en = English | es = Spanish | fr = French | de = German
| ja = Japanese.
Click Run Full Pipeline to execute the full multi-cloud sequence: Audio .wav → Speech-to-Text → Text Transcript → Language Translator → Multilingual Output.