The native Python SDK for SpeechWeave: background job polling, presigned uploads, and webhook verification. Python 3.10+.
Docs: speechweave.com/docs · API reference
pip install speechweaveSet your API key:
export SPEECHWEAVE_API_KEY="sk_..."from speechweave import SpeechWeave, wait_for_job
sw = SpeechWeave()
job = sw.jobs.create(
file="./podcast.mp3",
model="core",
service_mode="deferred",
)
done = wait_for_job(sw, job["id"])
print(done["transcript"])jobs.create accepts a local path string or an open binary file. For URL input, cancel, and other job operations, see the API reference.
Translate audio to English text, or fetch a completed job's transcript formatted as text, srt, vtt, or verbose_json (word/segment timestamps):
from speechweave import SpeechWeave, wait_for_job
sw = SpeechWeave()
job = sw.jobs.create(file="./spanish_podcast.mp3", task="translate")
done = wait_for_job(sw, job["id"])
print(done["transcript"]) # English text, regardless of the source language
# Once a job has completed, fetch its transcript in another format
srt = sw.get_job_formatted(job["id"], format="srt")When you already have an open file handle or in-memory bytes, use transcribe_file directly:
from speechweave import SpeechWeave, wait_for_job
sw = SpeechWeave()
with open("audio.wav", "rb") as f:
job = sw.transcribe_file(
f,
filename="audio.wav",
model="core",
language="en",
)
result = wait_for_job(sw, job["id"], timeout_sec=300)
print(result["transcript"])import asyncio
from speechweave import AsyncSpeechWeave, async_wait_for_job
async def main():
async with AsyncSpeechWeave() as sw:
job = await sw.jobs.create(
file="./podcast.mp3",
model="core",
service_mode="deferred",
)
done = await async_wait_for_job(sw, job["id"])
print(done["transcript"])
asyncio.run(main())from speechweave import verify_webhook
result = verify_webhook(
secret=WEBHOOK_SECRET,
raw_body=raw_body,
signature_header=signature_header,
)See the docs for a full FastAPI example.
from speechweave import SpeechWeave, SpeechWeaveError
try:
client = SpeechWeave(api_key="bad_key")
client.get_job("job_123")
except SpeechWeaveError as e:
print(e.status)
print(e.code)
print(e.error_type) # OpenAI-style category, e.g. "insufficient_quota"
# Prepaid wallet / spend caps: HTTP 402 with codes like INSUFFICIENT_BALANCE,
# WALLET_EMPTY, USER_SPEND_CAP_REACHED, CHECKOUT_REQUIRED, PLATFORM_SPEND_CAP_REACHED.
if e.status == 402 and e.code == "PLATFORM_SPEND_CAP_REACHED":
print("Monthly account limit reached; do not retry until next month.")
elif e.status == 402:
print("Top up the wallet or raise spend caps, then retry.")
# HTTP 403 with code EMAIL_UNVERIFIED: the account owning this API key hasn't
# verified its email yet. Verify it, then retry; the key itself is still valid.
elif e.status == 403 and e.code == "EMAIL_UNVERIFIED":
print("Verify the account email before uploading or creating jobs.")api_key, or setSPEECHWEAVE_API_KEYbase_url, defaults tohttps://api.speechweave.com/v1timeout, httpx timeout in seconds (default120)
If you are building a new application, use the native SDK above for full feature support. If you have an existing OpenAI, Deepgram, or AssemblyAI codebase, use the options below to switch with minimal changes.
Convenience helpers if you want OpenAI/Deepgram/AssemblyAI response shapes without adding another package. They use presigned uploads like the native API.
from speechweave import SpeechWeave
client = SpeechWeave()
with open("clip.mp3", "rb") as f:
result = client.audio.transcriptions.create(
file=f,
filename="clip.mp3",
model="core",
)
print(result["text"])Uploads go straight to storage the same way jobs.create does, so this supports files up to the same 250 MB self-serve limit.
More examples: OpenAI · Deepgram · AssemblyAI
You don't need this SDK for a quick swap, use the official openai package and point it at SpeechWeave:
from openai import OpenAI
client = OpenAI(
api_key="sk_live_...",
base_url="https://api.speechweave.com/v1",
)
with open("clip.mp3", "rb") as f:
result = client.audio.transcriptions.create(model="core", file=f)
print(result.text)client.audio.translations.create(model="core", file=f) works the same way for translating audio into English text; OpenAI's translations endpoint has no language parameter, the source language is always auto-detected.
Upload size: this path posts through the same wire format as the official OpenAI client, so it's capped at 90 MB per file to stay under standard upload limits. For anything larger, switch to
client.audio.transcriptions.create(...)from this SDK's drop-in helpers above: same call shape, and it unlocks the full 250 MB limit because uploads go straight to storage instead.
OpenAI model names like whisper-1 are aliased to core on our backend. See the OpenAI migration guide.